On March 26, 2013, Oracle announced a server refresh based on the new SPARC T5 processor[1]. The press release proclaims SPARC T5 is the “World’s Fastest Microprocessor”—an assertion backed up with a list of several recent benchmark results included a published TPC-C result.
This article focuses on the recent SPARC T5 TPC-C result–a single-system world record that demonstrated extreme throughput. The SPARC T5 result bested the prior non-clustered Oracle Database result by 69%! To be fair, that was 69% better than a server based an Intel Xeon E7 processor slated to be obsolete this year (with the release of Ivy Bridge-EX). Nonetheless, throughput is throughput and throughput is all that matters, isn’t it?
What Costs Is What Matters
There are several ways to license Oracle Database. Putting aside low-end user-count license models and database editions other than Enterprise Edition leaves the most common license model which is based on per-processor licensing.
To layman, and seasoned veteran alike, mastering Oracle licensing is a difficult task. In fact, Oracle goes so far as to publish a Software Investment Guide[2] that spells out the necessity for licensees to identify personnel within their organization responsible for coping with license compliance. Nonetheless, there are some simple licensing principles that play a significant role in understanding the relevance of any microprocessor being anointed the “fastest in the world.”
One would naturally presume “fastest” connotes cost savings when dealing with microprocessors. Deploying faster processors usually should mean fewer are needed thus yielding cost savings spanning datacenter physical and environmental savings as well as reduced per-processor licensing. Should, that is.
What is a Processor?
Oracle’s Software Investment Guide covers the various licensing models available to customers. Under the heading “Processor Metric” Oracle offers several situations where licensing by the processor is beneficial. The guide goes on to state:
The number of required licenses shall be determined by multiplying the total number of cores of the processor by a core processor licensing factor specified on the Oracle Processor Core Factor Table
As this quoted information suggests, the matter isn’t as simple as counting the number of processor “sockets” in a server. Oracle understands that more powerful processors allow their customers to achieve more throughput per core. So, Oracle could stand to lose a lot of revenue if per-core software licensing did not factor in the different performance characteristics of modern processors. In short, Oracle is compelled to charge more for faster processors.
As the Software Investment Guide states, one must consult the Oracle Processor Core Factor Table[3] in order to determine list price for a specific processor. The Oracle Processor Core Factor Table has a two-columns—one for the processor make and model and the other for the Licensing Factor. Multiplying the Licensing Factor times the number of processor cores produces list price for Oracle software.
The Oracle Processor Core Factor Table is occasionally updated to reflect new processors that come into the marketplace. For example, the table was updated on October 2, 2010, September 6, 2011 and again on March 26, 2013 to correspond with the availability of Oracle’s T3, T4 and T5 processor respectively. As per the table, the T3 processor was assigned a Licensing Factor of .25 whereas the T4 and T5 are recognized as being more powerful and thus assigned a .5 factor. This means, of course, that any customer who migrated from T3 to T4 had to ante-up for higher-cost software—unless, of course, the T4 allowed the customer to reduce the number of cores in the deployment by 50%.
The World’s Fastest Microprocessor
According to dictionary definition, something that is deemed fast is a) characterized by quick motion, b) moving rapidly and/or c) taking a comparatively short time. None of these definitions imply throughput as we know it in the computer science world. In information processing, fast is all about latency whether service times for transactions or underlying processing associated with transactions such as memory latency.
The TPC-C specification stipulates that transaction response times are to be audited along with throughput. The most important transaction is, of course, New Order. That said, the response time of transactions on a multi-processing computer have little bearing on transaction throughput. This fact is clearly evident in published TPC-C results as will be revealed later in this article.
Figure 1 shows the New Order 90th-percentile response times for the three most recently published Oracle Database 11g TPC-C results[4]. Included in the chart is a depiction of Oracle’s SPARC T5 demonstrating an admirable 13% improvement in New Order response times compared to current[5] Intel two-socket Xeon server technology. That is somewhat fast. On the contrary, however, one year—to the day—before Oracle published the SPARC T5 result, Intel’s Xeon E7 processors exhibited 46% faster New Order response times than the SPARC T5. Now that, is fast.
Figure 1: Comparing Oracle Database TPC-C Transaction Response Times. Various Platforms. Smaller is better.
Cost Is Still All That Matters
According to the Oracle Technology Global Price List dated March 15, 2013[6], Oracle Database Enterprise Edition with Real Application Clusters and Partitioning has a list price of USD $82,000 “per processor.” As explained above in this article, one must apply the processor core factor to get to the real list price for a given platform. It so happens that all three of the processors spoken of in Figure 1 have been assessed a core factor of .5 by Oracle. While all three of these processors are on par in the core factor category, they have have vastly different numbers of cores per socket. Moreover, the servers used in these three benchmarks had socket-counts ranging from 2 to 8. To that end, the SPARC T5 server had 128 cores, the Intel Xeon E7-8870 server had 80 cores and the Intel Xeon E5-2690 server had 16 cores.
Performance Per Oracle License
Given the core counts, license factor and throughput achieved for the three TPC-C benchmarks discussed in the previous section of this article, one can easily calculate the all-important performance-per-license attributes of each of the servers. Figure 2 presents TPC-C throughput per core and per Oracle license in a side-by-side visualization for the three recent TPC-C results.
Figure 2: Comparing Oracle Database TPC-C Performance per-core and per-license. Bigger is better.
The Importance of Response Times
In order to appreciate the rightful importance of response time in characterizing platform performance, consider the information presented in Figure 3. Figure 3 divides response time into TPC-C performance per core. Since the core factor is the same for each of these processors this is essentially weighing response time against license cost.
To add some historical perspective, Figure 3 also includes an Oracle Database 11g published TPC-C result[7] from June 2008 using Intel’s Xeon 5400 family of processors which produced 20,271 TpmC/core and .2 seconds New Order response times. It is important to point out that the core factor has always been .5 for Xeon processors. As Figure 3 shows, SPARC T5 outperforms the 2008-era result by about 35%. On the other hand, the Intel two-socket Xeon E5 result delivers 31% better results in this type of performance assessment. Finally, the Intel 8-socket Xeon E7 result outperformed SPARC T5 by 76%. If customers care about both response time and cost these are important data points.
Figure 3: Performance Per Core weighted by Transaction Response Times. Bigger Is Better.
Parting Thoughts
I accept the fact that there are many reasons for Oracle customers to remain on SPARC/Solaris—the most significant being postponing the effort of migrating to Intel-based servers. I would argue, however, that such a decision amounts to postponing the inevitable. That is my opinion, true, but countless Oracle shops made that move during the decade-long decline of Sun Microsystems market share. In fact, Oracle strongly marketed Intel servers running Real Application Clusters attached to conventional storage (mostly sourced from EMC) as a viable alternative to Oracle on Solaris.
I don’t speak lightly of the difficulty in moving off of SPARC/Solaris. In fact, I am very sympathetic of the difficulty such a task entails. What I can’t detail, in this blog entry, is a comparison between re-platforming from dilapidated SPARC servers and storage to something 21st-century—such as a converged infrastructure platform like VCE. It all seems like a pay-now or pay-later situation to me. Maybe readers with a 5-year vision for their datacenter can detail for us why one would want to gamble on the SPARC roadmap.
References:
[1] http://www.oracle.com/us/corporate/press/1923343
[4] Oracle SPARC T5 3/26/2013 http://www.tpc.org/1792, Intel Xeon E5-2690 http://www.tpc.org/1789, Intel Xeon E7-8870 http://www.tpc.org/1787
[5] As of the production date of this article, 2013 is the release target for the Ivy Bridge-EP 22nm die shrink next-generation improvement in Intel’s Xeon E5 family
Kevin Closson's Blog: Platforms, Databases and Storage 
Reality @ my work: SPARC is a vestigial organ in my data center, one that will be slowly extracted over then next 3 years.
I’m going to provide a devils advocate view just to stir you up a little:
The future of traditional enterprise data center computing is not so clear as your blog implies. Intel funded the development of Xeon microprocessors using their desktop parts. The difference between a Sandy Bridge Core i5 and a E5 Chip is not drastic and I think they share a execution core design. Macro-economically this doesn’t looks to be sustainable model as the desktop is a declining market. Just take a look at Intel’s key partners (Microsoft, HP and Dell). Intel’s operational revenue in this market is going to continue to shrink and they’re going to need to re-focus on ultra low power designs (cell phones, tablets) that greatly optimize power usage often at the cost of core performance. These designs will not adapt to become “Fat” processors for running traditional enterprise-y databases (aka Oracle). Thus “Fat” processors (Xeon E7, Power and SPARC) are going to become designs purpose built for a niche market with less shared engineering resources with desktop CPUs. This will be a a niche that Oracle will know how to play in exceptionally well. If you compare the where Oracle has come from (T2 & SPARC VI) to now they’ve made a huge improvement. Honestly I didn’t think Oracle had the engineers, tools and skill to close what was really a huge gap so quickly. I’ll be interested to see how this plays out.
Back to reality:
If I worked in a big “Oracle” shop (running lots of Oracle ERP on Oracle Database) and was still a big SPARC customer. At this point I’d say the course and just let this play out. I’d already have survived the darkest period anyway (When choices were SPARC64 VII and T2 ouch). What do I have to lose?
@Alexander , thanks for your feedback. So you’re saying that the strategic decision has been made in your IT organization to migrate off of SPARC over the next 3 years. Does that mean you’ll take on some T5 in the interim or have the brakes been put on and the re-platforming will just naturally take 3 years?
As for your ponderings of Intel’s future vis a vis server Xeons I must say that you have some interesting observations. If it didn’t so happen to be the case that I (under NDA) am privy to Intel’s roadmap that extends beyond the next 4 years I’d start to think you might be on to something 🙂
Its unlikely we’ll purchase any T5 servers, but you never know. We’re not big Oracle customers most of our workloads come from another bloated software company and run mostly on not SPARC. Our SPARC usage is mostly a hold over from some applications designed and implemented around the dot-com era, niche ISV’s, and some internally developed software. Really its more of a sideshow then a core data center platform. Sun was always a wildcard and although they couldn’t deliver on CPU designs they always kept us interested enough not force a trash bagging of the whole environment, it was basically winding down slowly as Sun did. The Oracle takeover killed all the fun and definitely has focused the products towards the “red stack” which we really don’t run much of so we’re pretty much done with it.
As the the road map I think the best approach is to evaluate whats in the truck now, look at your current situation and make a decision. I’ve seen enough failed and delayed road maps to know they’re not to be overly trusted.
@Alexander : your point is ***well*** taken vis a vis roadmap. That’s my point in my closing words. I wouldn’t gamble my fingernail clippings on SPARC roadmap. On the other hand, Haswell is essentially “on the truck” and it is no slouch.
I think with the latest announcements in the ability to migrate “on the fly” from SPARC based Oracle DB’s to Intel based Oracle with the big endian to little endian conversion being handled for you. To that end you can certainly choose to stay on big-iron UNIX for whatever your reason may be or choose to move to “commodity” where the price/performance is to your liking. You could go the HP way and get stuck on Itanic 🙂
Oh reality here is mixed. While some applications (not many) still run inside those big SPARC and the price for platform migration is so high that changing it is not a viable option. Even management will not allow a bold discussion on that because they know that the price for migration and their complexity will led to “unknow consequences”, so they throw the money to Oracle.
Meanwhile others applications (Oracle databases too) are migrating to Intel and Linux combo. Why? Price, price and price.
So the rule of thumb is: If the application is platform agnostic and it is able to be moved with minimal effort, the move is done, and the default is Intel.
Still SPARC will last for many years to come here, since new T4(1-4) arrived last month. SPARC will be a niche, a very expensive niche and Oracle knows that. Still surprised how core factor is just 0.5 for the new T5
@Icmarques : T5 core factor is .5 because it is only barely competitive with 32nm-process Xeon E7. If T5 were indeed the World’s Fastest Microprocessor then one would expect a core factor of 1 (as is the case with IBM Power).
@Alexander:
‘Intel funded the development of Xeon microprocessors using their desktop parts.’
That was the case during the Pentium 4 days and prior to that but in the post ‘Core’ era there’s no longer such thing as an Intel Desktop CPU. They make server CPUs and mobile/laptop CPUs. The desktop CPU line up now exist solely exists in the area of overlap between server CPUs being priced downwards and laptop CPUs being priced upwards. So the rationalisation of resources you mentioned has happened years ago but it was the desktop that was the loser, not the server.
We’re years away from ARM for server reaching critical mass and the old RISC UNIX segment (AIX on POWER, HP-UX on Itanium, Solaris on SPARC) will continue to wither in the face of Linux and x86.
The most Oracle can hope for is they transfer market share from HP-UX to Solaris. I don’t see this as a growing market I see it as a fight between Oracle & IBM for a steadily shrinking but very high margin business. AIX on POWER delivers IBM Mainframe style revenues and Oracle want a slice of that but it will never be mass market.
The laptop/desktop sandy bridge CPU core/pipeline is just tweaked to make Xeon E5, its not a redesign. More tweaking happens and out pops the sandy bridge E7 variant. They release the Desktop/Laptop parts first because they need the revenue to fund their business, if the server designs created more revenue they would release them first (Exhibit AMD). We can split hairs on the design origin, but its easy to see what hits the market first and where the revenue comes from.
Intel still sells a lot of laptop/desktop chips, but you can see their partners getting squeezed in the market by Samsung and Apple. Eventually it will blow back to them unless their mobile business takes hold.
>The laptop/desktop sandy bridge CPU core/pipeline is just tweaked to make Xeon E5, its not a redesign. More tweaking happens and out pops the sandy bridge E7 variant.
@Alexander : I’m sorry but that is quite incorrect…unless “tweaking” includes different pin count (2011 v 1567) and die size. E5 is 5.25 x 4.5cm and E7 is 5.65 x 4.917cm. In short, these CPUs have *very* little in common. In fact, Xeon E5 and E7 aren’t even the same microarchitecture. There was no Sandy Bridge E7. We are awaiting Ivy Bridge-EX (aka E7 V2).
I think Alexander meant that the execution pipeline, registers, L1/L2 cache, decoders, branch predictor, re-order buffers and anything else which is related to out-of-order execution engine is exactly the same between desktop and server CPUs.
@Alex : and he would be largely correct in saying so. That has little to do with the following words: ” More tweaking happens and out pops the sandy bridge E7″
I, and most everyone else, know the similarities between desktop and server SKUs. Saying E5 begat E7 through tweaks is the topic at hand…and an absurdity. Not to mention there was no Sandy Bridge E7. E7 is Westmere-EX.
Perhaps the word “tweaking” is a wrong one as it implies something relatively insignificant and that’s obviously not what’s happening to the server parts.
What Intel defines as core architecture stays the same between desktop and server parts. E7 including. Just because the package gets more pins, more cache, more cores and more yadda does not change the fact that the CPU core architecture between desktops and servers remains exactly the same.
All the difference between desktop and server parts comes from what Intel calls the un-core and system architecture but not the core itself.
@Alex Fatkulin: Yes, “tweaking” was the wrong choice of words. Then the original commenter mistake about E5 bagat E7 whether by tweak, or act of God, is just simply not a reality. Additionally I think you meant to type “microarchitecture” not “core architecture” but I don’t aim to put words in people’s mouths. I merely intend to respond to the words they write in their comments. So I think, maybe, you and I are now in agreement that there is more difference between Xeon E5 and the (non-existent) “sandy bridge E7 variant” than a could possibly come from a “tweak.”
All that aside, this post is about SPARC T5 so I’ll look for more comments regarding that. I think we are done with the insinuation that server Xeons are just polished desktop stuff.
Ok, “tweak” was a bad choice of words.
We’ve gotten off topic, but the goal of my statement was to say that I think common micro architecture between desk side and server CPUs is Intels business advantage vs. Oracle.
Also, I think @Storagezilla missed my point. Micro architecture used to build laptop and desktop CPUs is later used as building blocks to build server CPUs, not the other way around. “desktop” is just a convenient term perhaps I should have said “deskside” meaning CPUs in consumer laptops & desktops.
@Alexander J. Maidak : “common micro architecture between desk side and server CPUs is Intels business advantage vs. Oracle. ” Very well said, Alexander. However, I think it could be stated quite simply as SPARC has no volume play. It is, quite simply, a zombie.
That is the most significant point and it is, though indirectly, relevant to T5. If you don’t have a huge volume to spread your R&D investment over (as Intel does) the only other choice is to include this cost as part of the whole system and let it eat it.
As John Fowler said “we want to use non-proprietary memory so we can meet the cost targets”… oh wait a minute…
I also struggle to connect the dots on the Oracle strategy.
Best I can tell the goal is to have a very profitable zombie something like what IBM now has with the z Systems (Mainframe). Oracle’s leverage is the huge install base of ERP systems and custom applications written specifically for the Oracle DB.
The thing is they’re struggling in the Unix market and it seems to me that the workloads are more portable then the COBOL monstrosities created in the 1980’s and 1990’s.
Assuming I’m wrong about the application portability the only future state I see them succeeding is where Intel’s $34B PC business declines badly enough over the next couple of years that it can’t generate enough revenue to cover the capital invested in fabs and fabrication technology. Oracle doesn’t have this exposure and will be playing in a specific niche in which they mostly control the core software.
@Alexander J Maidak: I’m much more bullish on IBM than you. Google “transitive” or I offer this as food for thought : https://kevinclosson.wordpress.com/category/transitive/
You’re misreading how competitive laptop and above still is and assuming it’s where the bread is buttered. It’s not. Intel’s highest margin parts are the server processors as they have no viable x86 competitors. AMD’s server business is less a business and more a hobby for the fractional revenue it generates.
Every SKU Intel have under the Xeon line suffers from a level of AMD pricing pressure meaning regardless of lagging AMD integer performance there’s only so much Intel can charge for those parts before the pricing difference starts boxing them out and letting AMD along with it’s integrated & discrete GPU offerings into laptop and above wins.
Samsung and Apple are a moot point, Intel lost that war long before they sold XScale off in 2006 and it’s not coming back.
Swinging this back to Oracle. The SPARC T5 and trying to pull the plug on HP-UX support is all about trying to pick off the weakest member of the RISC UNIX herd (No pun on Hurd, well maybe) and that’s HP-UX. HP jumped from PA/RISC to the wrong Intel processor and have reached an evolutionary dead end. Oracle needs to grab as many of those Oracle on HP-UX customers as quickly as possible before they start losing them.
The bottom line is no one is building hyperscale datacentres or the next great app as a service on POWER, SPARC or whatever isn’t x86.
Anyone happy to buy all their Apps from IBM or Oracle (or SAP but not if you want HANA) should feel pretty good buying POWER or SPARC systems, everyone else is buying x86 and will continue to do so.
POWER & SPARC are the land that choice forgot.
@storagezilla : Threading on this WordPress theme is weak. Which comment are you replying to?
If your “niche” is all the big company big databases, that’s an OK problem to have. Over the next 5 years, there really isn’t a whole lot to slice off of that with the Big Data hype, and Oracle is covering that too (how successfully remains to be seen, regardless of opinions of how “well”).
Moving down market is where you see the Intel/linux, and it is really competing with Intel/MS. Make some popcorn, sit back and watch, but in the end it will be kind of like watching Greek wrestling, more standing around and grunting than eye-poking excitement. Your post makes an interesting point here, Oracle’s way is expensive in a price-sensitive area. But even more sensitive is the perception of the apps available, and that brings up a strong suit of Oracle Marketing.
Since Alexander mentioned the low power market, Qualcomm is eating Intel’s lunch, and it remains to be seen if Intel can deal with that. Or Apple, for that matter.
The Cloud hype will also affect this. It’s a good analogue of leasing, it’s more expensive, but it provides a service that is beyond the business that the lessee is in, handling hidden costs. But like timesharing and all the similar service models that came before, it will wind up suffering from margin erosion over time, as the back-end hardware and software mature. Bespoke back-end solutions will suffer the most if they don’t adapt. The largest hardware/software suppliers with the broadest offerings will win.
What I don’t get in the whole “licensing factor” rigmarole from Oracle is this:
if T3 is the slowest and gets .25 and T5 is the fastest and gets .75, why is it that P7+ gets 1?
Ah yes: because 2 and 2 isn’t 4, just 22…
😉
Hi Noons: Always a pleasure to see you here. Actually, it was that three-legged pink elephant UltraSPARC T2 that was doomed with the .75 core factor. T5 has a .5 which means Oracle’s money purse thinks it is on par with Xeon. The only remaining oddity is Itanium at 1 because Xeons are faster than Itanium. Power is charged at a factor of 1 because it *is* fast.
First, I’d like to point out how much attention Oracle is getting with its latest SPARC T5, SPARC M5 and even SPARC M10 announcements-Why?? if its so far behind? Even x-Oracle employees, EMC employees, IBM employees and Intel employees have dedicated lots of attention as seen here and in 100’s of articles. Its probably due to how amazingly fast Oracle has turned around the SPARC architecture from a laggard just 3-4 years ago to a leader today, even amongst the best of Xeon and Power.
Sure, its possible to cherry pick different performance metrics amongst the 3 different XEON architectures today (Westmere-EX, SandyBridge-EP, IvyBridge), as is pointed out here, but in comparing like for like architectures, meaning 8-CPU to 8-CPU, with same level of RAS, there is no faster throughput, faster/core, better price/performance architecture available today than the SPARC T5. Oracle has even fully disclosed all the leadership benchmark configurations here: https://blogs.oracle.com/BestPerf/
Regarding TPC-C, its just one of 17 world record benchmarks Oracle has published on SPARC T5 to help customers realize the performance of SPARC T5, and its systems capabilities, but you should never base conclusions on just one benchmark unless of course your workload is very closely related to the benchmark.
For the TPC-C benchmark, its clear that there are 2 primary goals. To process the maximum possible # of transactions and how much did it cost (over 3 years) to generate that performance. Yes, there are sub-benchmark details like response times that also matter, but bottom line is price/performance. Whats great is that all of the data is fully disclosed and must actually be approved by all vendors before being released. So its hard to do any tricks or do special configuration or pricing without being called out.
As for cherry picking different details within the TPC-C benchmark to declare Xeon leadership, the challenge is that you cant merge these 3 different systems into one to compare like for like, 1:1 against the SPARC T5-8 benchmarked. And it will be difficult at best to find any one x86 system that will beat SPARC T5-8 on all metrics.
Yes, the SPARC T5-8 outperforms the Oracle X4800/X2-8 8-socket Xeon system by 69% in throughput, but more importantly, 38% better Price/performance and even 6% faster performance/core. Yes, Intel Ivy Bridge-EX is expected at the end of this year (after a long 3 year wait from current/previous Westmere-EX) and we shall see what the improvements are in 6-9+ months. But if you look at the current fastest 2-socket and 4-socket systems, the generational improvements have been less than 15% clock to clock, so Intel will have to realize significant 69% gains to match SPARC T5-8 today. And clearly Oracle isn’t standing by either and SPARC T6 is on the roadmap for late 2014 with an additional 2x in throughput and 1.5x in single thread strength over SPARC T5.
“The TPC-C specification stipulates that transaction response times are to be audited along with throughput. The most important transaction is, of course, New Order. That said, the response time of transactions on a multi-processing computer have little bearing on transaction throughput. …shows the New Order 90th-percentile response times”
Well actually, we prefer to compare New Order *average* response times instead of 90th-percentile since that will show best and worst cases all averaged out. And comparing SPARC T5 to X2-8, SPARC T5-8 is within very respectable response times, especially considering the SPARC T5-8 offers 1.7x throughput improvement along with savings in price/performance. SPARC T5-8 also has over 3x faster response times than the fastest Power7 system benchmarked.
Comparing SPARC T5-8 to the 2-socket and 4-socket Xeon architectures that have much lower throughput scalability, the SPARC T5-8 has better response times. Go figure.
“Given the core counts, license factor and throughput achieved for the three TPC-C benchmarks discussed in the previous section of this article, one can easily calculate the all-important performance-per-license attributes of each of the servers.”
While many would like to just take the per core performance calculation into consideration to calculate the license costs, its important to understand how much throughput you can also get out of those licenses. Plus which systems are we comparing here? Notice that as Xeon scales, its performance/core significantly drops. If you need high throughput, you cant get highest performance/core.
So while the 2-socket and 4-socket Xeon systems have slightly better performance/core, they are considerably less powerful systems. After all, the SPARC T5-8 provides 1.8x more transactions/min for just 12% more licenses required.
Also, its clear that Oracle has advantaged SPARC for Oracle licensing over Power and Itanium (and same as XEON) because its goal is to sell OracleSW on OracleHW (and crush IBM and HP in the process)
Finally, make sure to look at the discounts used in disclosures as well, since different vendors have discounted differently on the benchmark! Cisco discounted 57% on their x86 system while Oracle discounted just 32% on SPARC T5-8.
Also, if you look at the breakdown in all the costs, you will see that the SPARC T5-8 total costs are roughly the same as the Oracle X2-8, but again, offering 69% faster throughput. The SPARC T5-8 HW costs are actually very similar to the 40-core IBM x3850 system with 1TB less!
And so finally, as I said in beginning, its important to look across several benchmarks to get a good idea on “average” performance, as well as performance/CPU, performance/core and ultimtaley price/performance. As you’ll see, SPARC T5 is very competitive. Ive attached a presentation highlighting all the remaining benchmarks and performance results we’ve achieved with the SPARC T5 systems.
And yes I disclose I work for Oracle, but my comments are all my own and may not reflect those of Oracle’s.
@Phil Dunn wrote:
“And yes I disclose I work for Oracle, but my comments are all my own and may not reflect those of Oracle’s.”
No problem. You are welcome to comment here. I have a few counterpoints/questions:
1) You wrote a lot about how Oracle Database on SPARC slams Oracle Database on Oracle’s own manufactured Xeon-based servers. Why does Oracle sell Xeon-based servers?
2) You wrote “Oracle has advantaged SPARC for Oracle licensing over Power and Itanium (and same as XEON) because its goal is to sell OracleSW on OracleHW (and crush IBM and HP in the process).” May I ask whether you think Oracle would be able to employ tactics if not for their dominant market position in commercial RDBMS?
So are you agreeing with my points? I don’t really see any counterpoints here.
1) Well, just like Intel offers different choices of Xeon and even Itanium architectures, IBM offers both Power and x86 and Oracle offers choice between x86 and SPARC. Before Oracle acquired Sun, it was all x86 and today, Oracle continues to have a two horse race, although its money is predominantly on SPARC today. The challenge that Oracle, like IBM has is that Intel controls direction of Xeon, not Oracle or IBM as so theres less opportunity to fully engineer Xeon for Oracles or IBMs specific software.
2) Just like IBM has its PVU licensing policy and its designed to incent customers to run IBM SW on IBM hardware, Oracle has its SW policy and is focused on selling Oracle on Oracle. While Oracle clearly has a dominant position in RDBMS, its not # in all other categories of SW where it has the same multipliers so what is the tactic here? And anyway, whats wrong with Oracle making its software more (financially) attractive to run on its hardware?
@Phil_Oracle : No, we don’t agree on much but I have let you have your say nonetheless. You wrote: “Oracle clearly has a dominant position in RDBMS” followed by “whats wrong with Oracle making its software more (financially) attractive to run on its hardware” to which I would reply a) depends on who you ask and b) depends on the whether the who you happen to ask happens to be an antitrust attorney. Pass it by one of them and let me know what they say.
DISCLAIMER: I’ve merely done some leisure-reading on such matters as antitrust law. I make no claims of expertise in such matters.
Oh, what do you suggest is the 120 PVU per core value that IBM is using for the T5-8 (same as for 770, 780 and 795 😉 )?
@c0t0d0s0 : Sorry, Joerg, I don’t understand your point. Can you say more on the matter please?
I ask again with some more information: PVU is the licensing metric of IBM, Works a litte bit differently than the multiplicator, however has the same effect: A PVU has a certain price for a product, to know how many PVU you need, you multiply the number of cores with the PVU value of a core of a certain kind (there is a list) and multiply this with the price of a PVU. One core in the p780,p770 and p795 has the same per core value as a core in a T5-8. What does this suggest to you?
@c0t0d0s0 : Is this PVU stuff for DB2 pricing? Sorry, I’m not aware of this stuff.
Reply to a the answer of Kevin to my answer: Just to follow your arguments … perhaps out of the reason because someone at IBM thinks T5 is really fast 😉
@c0t0d0s0 “Reply to a the answer of Kevin to my answer: Just to follow your arguments … perhaps out of the reason because someone at IBM thinks T5 is really fast ” They take Oracle’s word for it but Oracle doesn’t? T5 is .5 Power processors are 1. The irony on one side doesn’t cancel out the irony on the other side. Nonetheless, I still have no idea what that PVU is about. Is that how IBM prices DB2 ?
yup.
@c0t0d0s0 : So I fail to see your “smoking gun.” DB2 pricing: same whether Power/AIX or SPARC T5. Oracle: 50% less expensive on T5 (compared to Power/AIX) even though Oracle deems T5 the World’s Fastest Microprocessor. Please explain to us how your logic helps your “cause.”
I should add a link to this relevant reader comment: https://kevinclosson.wordpress.com/2012/03/22/oracle-database-11g-certification-for-rhel6-oel6-has-finally-materialized/#comment-38387
Sure, if you are comparing performance / cost then there might be other better alternatives. But Oracle is talking about fastest performance / cpu, not performance / cost. I agree with you that if you factor in performance / cost, then the Oracle benchmark is not relevant. But if you are only considering fastest performance, then Oracle benchmark is relevant.
Best performance/cpu? Or, best performance/cost? The choice is yours.
@Mads: You wrote: “Oracle is talking about fastest performance / cpu.” You’re right. The SPARC T5 demonstrates 13% faster response time than a 2U pizza box. Take a look again at Figure 1 above to see the comparison to nearly-outdated Xeon E7 technology (the E7 is ~2x faster).
Well, The SPARC T5-8 actually has better price/performance than *all* 8-socket systems on the TPC-C benchmark and actually beats many of the leading x86 systems as well. Its currently ranked #9 best price/performance and when you look at #1-#9, they come no where near the total throughput. http://www.tpc.org/tpcc/results/tpcc_price_perf_results.asp?resulttype=noncluster
@Phil_Oracle : I discredit pricing in TPC-C. They are just pricing tricks. Moreover, the SPARC T5 result didn’t use a storage product anyway
. Oracle used an erector set of cobbled COMSTAR-headed F40s. On the other hand I accept the host-level performance proof exhibited by TPC-C. Let me put that another way. No customer on the earth will ever cobble together the storage bill of materials specified in the SPARC T5, nor X2-8, result so it doesn’t matter what that component of the test configuration costs. Oracle could price that non-storage product at $1 for all any reasonable person would care because it is not a product, noboday will ever deploy it and it is downwind of the host so we don’t care. TPC-C is not a storage benchmark. It is a host benchmark–and a good one at that. This is why I turned it on it’s head and focused on core factor and throughput per list-price licenses dollar.
Finally, since your last comment doesn’t mention the word “fast” and instead used the word “throughput” we can now agree on something. The T5 is currently the single system throughput king. No interpretation necessary to draw that conclusion.
Oh, actually i personally have a very large customer exactly using this configuration. With success. And no … i can’t tell who it is. And: Solaris 11 is a product, the server is a product, the F40 is a product. What do you need more? Additional i used dozens of X4540 in such a function. It’s perhaps unusual not to use storage that comes out of the box read to plug-in. But that makes the storage used in the TPC-C benchmark not an non-viable choice. And at the end: What is highend storage at the end: A server with an OS, an FC/SAS/iSCSI device in target mode and some disks.
@ c0t0s0d0 : Ok, Joerg, you’re right. I should not have said “no customer on earth.” It’s true, sometimes man bites dog! 🙂 I wouldn’t ask who the customer is.
I do not argue your points about Solaris 11, nor the fact that the servers are a SKU. The F40 is indeed a product. It’s the LSI Nytro.
By the way, the SPARC T5 TPC-C used host volume management (SVM) instead of ASM. Is that a common thing to do when using the COMSTARstuff(tm)? Why do you suppose Oracle’s benchmark team left ASM out of the SPARC T5 TPC-C ?
@Phil. I suppose that with the great engineering success of the SPARC T-5 and the forth coming T-6 CPUs. Oracle will refresh the Exadata Storage and Database nodes from the those slow outdated Intel X-86 processors for the new SPARC gear…. maybe even in-time for OOW 2013. The new multi-threaded feature set in 12c will certainly enjoy all those threads in the new SPARC platform.
As for TPC-C, why not call me at Violin so we can run a TPC-C or TPC-E benchmark for you (like we have for HP, Cisco, Fujitsu, Dell running Oracle databases) so you don’t have to cobble together all of those Comstar heads and F40 devices that no-customer will ever by. An build a test platform that your customers will actually want use.
As for IBM – catch the wall street journal article, Big Blue’s x86 business is up for sale.
@Kevin. Another good article.
@Matthew Morris: Well, when Per Larsen sold the thinkpad division to lenovo folks thought he was crazy. I think Lenovo can do good things with all that Sequent IP IBM has in the xSeries line (e.g., eX5) 🙂 … oh, glad about the EMC Lenovo relationship too.
So, @Phil and @c0t0s0d0 insist that COMSTAR FrankenStorage(tm) is viable and that I’m cherry picking for not accepting the TpmC/$$ comparison of the aged E7 result compared to the T5 result. I have a follow up blog post coming on that but maybe the graphs I point to in the following tweet tell the story:
https://twitter.com/kevinclosson/status/325316996285751296
I love these ‘jihads’ 😉 All I can say is, we used to run a lot of databases on Intel/Linux and we consolidated all of them onto T4-4 servers. The I/O performance is fantastic. More importantly the stability & performance of Sol10 on T4-4 is mind boggling. When was the last time you logged into a server with a run queue of 1,000 on a Linux server and had it respond to your commands? Why did I’ve a run queue of 1,000? That’s another story with an incredibly poorly written SQL and lots of them in a COTS application.
As some one in an earlier post said, the fact that T4/T5 are even in discussions is a great sign. Intel’s ability to focus on the server market when they are getting killed in the mobile/smartphone, and not to mention Itanic, is to be seen. Intel is a formidable competitor and I’ve no doubt they’ll continue have nice and compelling products.
Instead of dinging Oracle, let’s enjoy the competition and benefit from the open markets regardless of our platform preference (I’ve mine, that is obvious and so is yours)
Anatha,
I’m sure you’re aware that you don’t need SPARC to run Solaris. We’ve been running for years on x64.
Today processor like Sparct T5 and new Xeon E7 Family are really fast.
There is no doubt about this.
The main issue is … downtime !!!
If your business requires no downtime to have only profit your answer is Sparc T5.
Do you need the best ? M5-32 is your choice.
Best Regards,
I’ve had a series of SPARC system processor boards fail in my experience and just because it’s expensive doesn’t mean it’s necessarily good. I’ve got plenty of x86/x64 systems that have uptimes of a number of years and no failures. A well architected system virtualized on Intel Xeon’s will outperform the T5 system any day of the week in any number of metrics that matter, including lower TCO, acquisition cost, operations and even performance in some cases. No point investing in a dying bread and being locked into ever increasing maintenance costs. Any application that can be moved will be moved and those that can’t be moved now will be moved at some point, it’s just a matter of time.
@vcdxnz001 : Thanks for that info. If folks would take the time to measure they’d find that single stream on T5 is quite wretched as well. Just run an import, or catalog.sql or anything that is a single CPU-burner and compare to even 115w E5 parts (e.g., 2665) and you will be sorely let down.
T5 is a bandwidth processor and a good one for that. It will enjoy being king of some oddly-defined hill for just a few months.
Kevin-Do you have any substantiation to justify your claim “single stream on T5 is quite wretched”? Have you actually tested a SPARC T5? Based on the 18 world record benchmarks published on SPARC T5 across pretty much all of the main Oracle software in every sector, SPARC T5 is absolutely outperforming our fastest x86 systems (and any competitor x86 systems) from price/performance, throughput, performance/core and even critical SLA response times. Have you seen this? A SPARC T5-2 outperforms an 8-socket HP Proliant DL980 G7 on critical throughput under service level agreements (SLAs) specifying response times.
https://blogs.oracle.com/jimlaurent/entry/solaris_11_provides_smooth_scalable. Unless you can provide the data to back up your statement, we’ll have to assume you haven’t.
Hi Phil,
I have a feed of data coming to me from the owner of a T5, yes. But, please read on.
Since I didn’t cite my numbers when I said “single stream on T5 is quite wretched” what you are quoting is my personal assessment based on analysis of non-disclosed data. I suppose it may be the case that you really think I have no data upon which to base my assessment. If so I would encourage you to just stop reading my blog. As I said, I do, in fact, have data but I am not at liberty to share that data. In the same comment where you extracted that sound bite I also wrote: “T5 is a bandwidth processor and a good one for that.” We all know, based on Oracle’s TPC-C results, that on a per-socket basis the T5 outruns any current Xeon. Those are audited/published results. Those results prove to us that the T5 is a huge bandwidth system. Do those sound like disparaging words to you? You, on the other hand, are trying to say T5 is the best in every manner of performance and characterization–to include all manners of price-performance–under all circumstances. I’m saying T5 is good in some ways, poor in others. Which do you suppose has more credibility?
If you have any proof point that shows a single T5 core outperforming a single current Xeon core please post it. Benchmarks run at 100% processor utilization. Production systems usually do not so single-stream performance is important as well. Single stream performance, by the way, shows up as increased transaction response times such as those of New Order in the published TPC-C results. I wrote about that in this post.
I gladly let you speak your mind here. In return I ask that you presume I have measurements that back my assessment. I am putting T5 into a good performing bucket–specifically bandwidth. That’s not such a bad place to be. Is it?
I will retract my assessment of my data if you can provide any single-stream performance data that counters it. That is, if you can provide any industry standard single-stream metric that counters the data I have I’ll gladly reverse my assessment and presume my data is somehow invalid. Alternatively, you could just ignore what I write because unless I provide citations of my data I am only sharing my opinion as specifically called out in my blog disclaimer.
“SPARC system processor boards”? Which systems are you talking about? Sounds like they are 5-10 years old and yet you insist that SPARC T5 is the same?
“A well architected system virtualized on Intel Xeon’s will outperform the T5 system any day of the week in any number of metrics” Its one thing to say whatever you want, its another thing to substantiate it. So far, you’ve been proven wrong by Oracles 18 world record results on SPARC T5 and Oracle sells both x86 and SPARC systems. https://blogs.oracle.com/BestPerf/
“No point investing in a dying bread and being locked into ever increasing maintenance costs” Have you been living under a rock these past two years? Dell is trying to go private to escape public scrutiny for its suffering (x86) business, HP’s (x86) hardware business is crumbling and IBM is trying to sell off its x86 business to Lenovo. Good thing Oracle didn’t bet 100% of its business in x86.
As for maintenance costs, you clearly haven’t looked at Oracles new Premiere Support (last ~3-years) for Oracle HW (both x86 and SPARC) where Oracle charges just 12% of hardware net price, fixed price, per year, which includes 24/7/2-hr response time for HW/SW support, OS & Virtualization licenses and support, System Management software licenses and support and Preventive Services. If you price out all of these things with any other vendor, regardless of OS and virtualization, I can almost guarantee it will be more expensive (especially with a lot of hidden post-sale charges).
philoracle — I’m just a curious observer on this blog but I can’t understand how can you claim “performance/core” advantages T5 has over Intel?
If you look at Oracle’s own data:
https://blogs.oracle.com/BestPerf/entry/20130326_sparc_t5_speccpu2006_rate
One processor results SPECint_rate2006 peak:
T5 gets 467 (16 cores)
Xeon E5-2690, 2.9 GHz gets 357 (8 cores)
so T5 has 28.5 per core and Xeon 44.5 per core — that’s a huge difference.
I think you can look at the number of threads and do the similar math too.
Keep in mind that increasing singe core/thread performance is not the same as stuffing more cores on the chip.
What do you think can happen faster — Intel stuffing 2x of (the same) cores onto the chip or Oracle figuring out how to increase per-core performance by 56% by the means of uarch improvements?
Well, the way I look at it, is if you’re doing performance/core calculations its probably because you’re analyzing licensing costs and therefore you should probably be looking at a benchmark that actually runs a similar type workload and clearly SPECint2006rate is NOT- Its pretty much a RAW CPU benchmark that doesn’t take anything else about the CPU nor the rest of the “stack” into consideration like system I/O, networking, OS, DB, encryption, etc. For example, a DB clearly needs massive I/O yet this benchmark avoids testing I/O? So, I’d recommend doing per core calcs based on one of the many application/database based benchmarks like TPC-C, TPC-H, SAP or one of the SPECj benchmarks which will highlight more realistic differences. Also, the SPECint2006rate benchmark is very HPC biased(many of the subtests include Artificial Intelligence, Quantum Computing, etc) which clearly is not Oracle’s R&D focus.
Another point is that you are comparing a Xeon E5-Series CPU which can only scale to 2-CPUs (16-cores Max) to the SPARC T5 which scales all the way up to 8-CPU’s (128-cores). So sure, at up to 16-cores Xeon may have a perf/core advantage, but clearly that perf/core advantage is lost when you look at a 4-CPU or 8-CPU Xeon vs T5 comparisons.
If you compare 8-CPU systems, you will see that even on SPECint2006rate – SPARC T5-8 beats DL980-8 in RAW CPU perf/core for example.
——————————————————————-
SPARC T5-8 w/ 128 x SPARC T5 @ 3.6GHz cores achieves 3,490
SPECint2006rate base = 27 SPEC/core
HP ProLiant DL980 G7 w/ 80 x Xeon E7-8870 @ 2.40GHz cores achieves SPECint2006rate base = 26 SPEC/core
If you look at all the other benchmarks, the ones with the higher I/O demands like TPC-C or TPC-H will show the best strengths of SPARC T5 vs Xeon, since the rest of the stack is being tested.
For example:
TPC-C – SPARC T5-8 beats Oracle X2-8 by 6% perf/core
SPECjEnterprise2010 – SPARC T5-8 beats Oracle X2-8 by 32% perf/core
SPECjbb2013 – SPARC T5-2 beats HP DL560 G8 by 12% perf/core
TPC-H (1TB) – SPARC T4-4 beats HP DL980-8 by 2.3x perf/core
TimesTen In-Memory Database – SPARC T5-2 beats Oracle x2-4 by 3x perf/core
And to answer your final question “What do you think can happen faster — Intel stuffing 2x of (the same) cores onto the chip or Oracle figuring out how to increase per-core performance by 56% by the means of uarch improvements?”
Well, its clear that almost all CPU developers are moving to more and more cores /CPU as more GHZ/cache isn’t enough and with every increase in # of cores, theres usually a decrease in perf/core. However for Oracle, as has been stated publicly, the future of SPARC is all about Software in Silicon. And where Intel may be able to achieve 15-20% performance increases generation over generation, Oracle is expecting 2x-5x increase in performance with future SPARC (running Oracle SW of course). This is something that Intel just cant do.
@philoracle : I appreciate your contributions to this thread. They will make for enjoyable reading by all over the next 2-5 years.
philoracle — the thing about SPECInt is it clearly shows that T5 has a big deficiency in IPC (Instructions Per Clock) compared to Intel (I will touch on Xeon E7 point a little later).
Let’s take a look at TPC-C numbers quoted by you:
T5-8 archived 8,552,523 tpmC or 66816 per core
X2-8 archived 5,055,888 tpmC or 63198 per core
so there is indeed 6% advantage per core but here is where the bad news are coming… Xeon E7-8870 is more than two years old and three generations (Sandby Bridge, Ivy Bridge, Haswell) behind.
What do you think is going to happen when Xeon E7 V2 based on Ivy Bridge will be released later this year?
You may say there is price/performance advantage with 0.55 for T5-8 vs 0.89 for X2-8. But run a quick numbers experiment — replace the storage used for X2-8 with storage used for T5-8 — see what happens?
I can go through the rest of your list in a similar way.
> so there is indeed 6% advantage per core but here is where the bad news are coming
Alex,
That 6% advantage in TpmC/c for T5 is crystal clear in Fig 2 of the blog post. Nothing you can say will make Phil accept any position other than T5 is the best, fastest, cheapest at all performance characterizations for all workloads under all conditions with the best price-performance to boot.
I have a part 2 on this blog series that draws out the fact that Oracle’s T5 price-perf is tainted by the fact that they assembled significantly less expensive storage. The shape of the pie is skewed to T5 due entirely to the fact that the COMSTAR erector set they attached to the E7 has tons of expensive F20s as opposed to cheaper faster F40s servicing I/O to the T5. Both cases though it was FrankenCOMSTAR(tm) non-product storage so I’ll shave the kit off above the storage in the next post.
Am I the only one that sees a complete conflict of purpose in putting CISC on RISC – aka “software on silicon”?
No wonder T5 cannot be pushed faster on individual IPC: CISC is not RISC, and a penalty must be paid for trying to make them coexist in same chip…
The other side of course is: general purpose computers (as opposed to db machines) do NOT just run rdbms code. In fact, db code is a small portion: OS, network, file system, etcetc, don’t need and won’t use Oracle on a chip, now or never.
The whole enchillada is completely flawed by design.
But let’s allow time to prove that one: it’s only an encore of tried and failed ideas, it won’t change now…
Nuno,
remember that x86 CPUs are essentially CISC, at least from the outside. The front-end converts all instructions into RISC-like internal uops which are easier to work on.
IBM is doing specialized co-processors with their zIIP and from what I’ve seen people would rather have a general purpose CPU which can run all workloads.
T5 appears to be a step backwards from T4 in terms of IPC though it was probably a conscious design decision to reach higher clocks. But on SPECjbb2013 T4 beats T5 clock-for-clock with a substantial margin.
So there we have it, a less efficient CPU but with higher clock speeds and twice as many cores crammed together. How often can you pull something like that considering the speed of lithography process improvements?
Did anybody see Max TDP for T5? Oracle had it in T4 datasheet and it mysteriously disappeared for T5.
Alex: I am fully aware and remembered of CISC and RISC CPU architectures – a lot more than you can imagine. Please recall that I worked 10 years with computer makers before I went to Oracle. And CISC and RISC are not exactly new CPU architectures. Even in this day and age of renaming “service bureaus” as “clouds”. 🙂
Commenting “CISC on RISC” for what Oracle is doing in T5 does not demean in any way what X86 or x64 do.
What x86/x64 are not is RISC. Not anymore than Sparc is CISC.
Regardless of Oracle’s marketing statements on the subject.
I fully agree on the general workloads. They are capable of being handled with CISC or RISC. But “or” is not the same as “and”. What is doubtful are the generic benefits of a franken-architecture such as Oracle claims for T5. And more than likely why so many are seeing inconsistent performance numbers. But as more details become known of exactly what has Oracle done to Sparc, it’ll all hopefully become a lot clearer.
For now, all I can say with absolute certainty is that special purpose macro-code instructions for handling Oracle rdbms NUMBER (a modified packed decimal) – and a few other published specifics – are totally useless in a general purpose computing model. Which is what the vast majority of computers out there are.
Useful perhaps for db-machines. But I’m also reminded of how most of the historical ventures in that space (Britton-Lee anyone?) went the way of the dodo. Let’s hope it won’t be the case here.
@Nuno : Good points. I’ll chime in. Oracle’s Marketecture futureware known as “Software in Silicon” is only futureware or better yet ideaware to buy mindshare amongst the pundits.
In the real world with legacy RDBMS, the instructions wasting cycles are not complex. They are, in fact, very simple instructions that stall on memory or waste cycles spinning on contentious memory–spinlocks and other such mutexes. It’s that way on all SMPs and always has been. Some systems have mitigated this by implementing hardware locks such as Sequent with NUMA-Q and Convex Exemplar (what’s inside a superdome) and now with Haswell (lock elision) , Intel Xeon.
Moving up the food chain of instructions immediately takes one to functions that manipulate shared pool and SGA buffers (kcb*). The instructions in these routines do burn cycles but not on-die. They burn cycles stalling for load/store instructions and no matter what Oracle’s supposed SPARC futureware Software on Silicon is supposed to do, it will not improve load/store. Those are not instruction costs but memory costs.
Once a line of memory is in the processor cache and the code is doing something like whacking an Oracle NUMBER data type well, yeah, sure, there could be some instructions to shave off if all of the code for manipulating packed decimal were a single CISC instruction. So the question becomes: what is more expensive the load/store or the instructions tickling the line when it is in processor-cache? The answer is silly-simple. Load/store is what costs.
For instance, once a cache line is in memory, even shaving off 90% of the instructions that manipulate it pale in comparison to the time the processor lost stalled on loading the line in the first place. Accelerating small fractions of a problem is not the way to go.
If Oracle wants to make the database run faster, um, well, er, they need to reduce the number of load/stores per transaction.
Hey Nuno,
You don’t seem to know NADA about Oracle & SPARC’s “Software in Silicon” strategy. Clearly you need to do some homework as Fujitsu/Oracle has been shipping the SPARC M10 based systems with “Software in Silicon” since January of this year http://bit.ly/1a90WWw and many of the technologies in the shipping S3 core based SPARC T5/SPARC M5/SPARC M6 processors would be considered part of this strategy including “dynamic threading” and “critical threads” which can give up to a 15% improvement in DB response times, 30% improvement for coherence and even a 2x speedup for SOA startup. Oracle just published a newer version of the SPARC T5-8 benchmark on SPECjbb2013 where it saw a 30% performance improvement just by upgrading to Oracle DB12c (and other minor config changes), which implements many of the on-chip technologies. Theres more to come, and if you’d like to educate yourself, look here: http://bit.ly/18KEkLa
Hey Phil,
I don’t think you fathom what I know about Oracle and Sparc SIS strategy. Suffice to say for the last year or so, Oracle has tried – completely, utterly unsuccessfully – to sell the company I work for, the M10 and other T5 hardware.
In the process we’ve been treated to umpteen presentations full of “powerpoint proof” of how backwards and antediluvian we are.
By contrast, Oracle’s competitors – and who have won the bid (I wonder why?) – started by asking us WHAT we do, in detail! HOW and HOW MUCH of it. And then provided a solution that fitted us to a “T” (pun intended), while providing plenty of room for growth.
A slightly different approach from twits at Oracle/Fujitsu/Sparc who fill conference rooms yelling at each other how good they are – while the rest of the world just laughs and walks the other way.
You see: there is a FUNDAMENTAL and BASIC difference between claiming a feature “can give” and delivering something that does give.
Oh, there isn’t? Then, may I suggest you improve your education before commenting on other’s?
Yes,yes, I know that Oracle and Sparc are planning to put Hadoop as a pluggable db in 12c – or whatever follows it. Completely irrelevant for the vast majority of database users out there. But do not allow simple market reality to stop you from doing so!
You see: Rick was here too, talking to us. And I didn’t need to join the “secret meeting” crowds at OOW to learn what the “future” is!
Ah yes: enjoy a few more quarters of missed sales targets. Then hopefully, it’ll (finally!) sink-in…
Apologies if I was a bit aggressive in my remarks as I see a lot of sarcastic and demeaning comments going on between you and Kevin related to Oracle. I should have directed some of my frustration towards Kevin especially remarks like “Oracle’s Marketecture futureware known as “Software in Silicon” is only futureware or better yet ideaware to buy mindshare amongst the pundits.”. Software in Silicon is a huge project at Oracle and clearly the Database engineers, as well as the entire Oracle SW engineering teams are working together with SPARC engineering to figure how to get the absolute best performance out of SPARC. 5 CPU generations in just 4 years is amazing. Kevins comments that SIS is future-ware/idealware is misleading. Maybe because EMC is 100% focused on x86? Fujitsu SPARC M10 has software in silicon today and seeing amazing results with 12c. We have next gen SPARC CPUs up in the labs that have SIS and will be out by next year. Too bad you didn’t chose Oracle HW (and Solaris). You wont be able to leverage any of this R&D including whats special in 12c. And BTW, sure, we can be aggressive at times, but its just because it’s the nature of the business-competition is fierce. And note, Oracles quarterly revenues continue to be positive while many of its key competitors are suffering like HP, IBM, and even Intel. My comments are all my own…
@Phil_Oracle :
The problem this the term “Software in Silicon” is it is simply a hyped term for what we already have. You aren’t flipping dip-switches, are you? If the SPARC guys want to cook a single CISC instruction that implements the entirety of kcb that’s perfectly fine with me. But in the end it is just a CISC instruction.
Google can tell you what kcb is.
@Phil_Oracle :
You wrote: ” You wont be able to leverage any of this R&D including whats special in 12c” to a long-time Oracle customer who pays 22% of list in perpetuity for software maintenance. As you speak of all of these “features” in the database be aware of who is funding the development of these features. If you don’t know it is probably because you are no lot of tooth Oracle and don’t know that dev is funded by maintenance revenue.
Apology accepted, I know perfectly well the kind of pressure the hardware business is in, having spent 13 years of my prior IT career in it.
The thing you and all the “engineers” don’t seem to realize is that NO ONE in his right mind will buy into an architecture that gets 5 generations in 4 years.
What that tells me is that there is an abyssal separation between the actual real market and what the Oracle hardware folks think it is.
Let me be very clear here: the fact that Fujitsu has that multitude of architectures in 4 years does NOT mean by any stretch of the imagination that the last one is going to be adopted by all those who got sucked into the first one. There is a FUNDAMENTAL difference between “having” and “selling”.
Yes, yes, we know Fujitsu and Sun have the entire market in their hands and no one else is selling anything – can I laugh now, or do I have to wait a bit longer?
Oh, and please: do NOT confuse Oracle’s bottom line as a company and its hardware division’s bottom line. Again, abyssal difference.
The Oracle hardware business continues to hitch a ride at the expense of mine and other’s software maintenance revenues. Take that away and the entire Sun business would have been ejected years ago as the loss-leading and loss-making piece it truly is. Careful, it’s not just me who can read a company’s annual prospectus and analyze it.
And again: do not confuse internal testing and benchmarks of 12c by folks who are charged and contractually obliged with saying whatever Oracle marketing wants, with actual installed user experience.
No one (outside of Oracle) in his right mind would use 12c for anything near production for the next 3 years, given the pitiful history of Oracle software stability in the past 20 years.
As for my liking or not of Oracle – the software, NOT the hardware! – I’ll leave you with a couple of thoughts:
I’ve been using it for 25 years, first at Oracle, then outside and I have yet to find a better piece of database software, once it’s been stabilised – that is why I keep using it and recommending it.
The place where I work would have moved to M$SQL years ago if it wasn’t for my efforts to keep Oracle here – we shove over a quarter of a million bucks up Larry’s bottom line as a result.
This is why the thanks I get from some twits inside that Sun company is to insult me whenever I disagree with anything. Let’s not go into the multiple unsuccessful attempts (once every 3 months for the last 7 years) by their consultancy bozos to get me replaced/outsourced by whoever is the fancy “partner” of the month…
Well, I can see you have quite a bit to say. And heres my point of view:
“The thing you and all the “engineers” don’t seem to realize is that NO ONE in his right mind will buy into an architecture that gets 5 generations in 4 years.”
From your comments you must be pro-Intel then (or maybe IBM Power?), but this doesn’t make sense since Intel has been releasing new CPU generations almost every 6-9 months and in same timeframe as Oracle/Sun, Intel has released 6/8 server oriented CPUs (depending on how you look at it) from Nehalem-EX (65xx), Nehalem-EX (75xx), Westmere-EX (E7-2xxx), Westmere-EX (E7-4xxx/8xxx), E3-12xx-series & E5-14xx/16xx/24xx/26xx/46xx-series “Sandy Bridge/-EP” to E3-12xx v2-& E5-16xx v2/24xx v2/26xx v2/46xx series Ivy Bridge/-EP. Doest seem to have impacted adoption of Xeon. More releases shows theres investments and advancement which to me are good things!
“What that tells me is that there is an abyssal separation between the actual real market and what the Oracle hardware folks think it is.”
What it tells me is that Oracle is serious about CPU development and is investing $BN’s in advancing SPARC (besides what Fujitsu is investing) to what was promised it would deliver in 2010 (via the roadmap) after Sun acquisition, and it has delivered to date. I don’t hear anyone complaining about poor performance anymore. Or high prices. Times have changed. Even the x86 market has dramatically changed, especially since ARM arrived in force. HP, Dell, Sony and even Intel are really hurting in revenues these days due to their significant dependence on x86, which has clearly become commoditized with razor thin margins which no one can survive long on.
“Let me be very clear here: the fact that Fujitsu has that multitude of architectures in 4 years does NOT mean by any stretch of the imagination that the last one is going to be adopted by all those who got sucked into the first one. There is a FUNDAMENTAL difference between “having” and “selling”.”
Well, looking at the analyst marketshare #’s, Fujitsu & Oracle SPARC did well over $1.5BN in revenue these last 4 quarters (in a financial crisis market no less), shipping over 45,000 servers. That’s roughly half of total Unix servers sold. And that’s not including 3000+ engineered systems sold with well over a $1BN run rate YoY per FY13. And we wont talk about 15+ years of binary compatibility that Solaris has had, allowing lots of legacy code to continue running on latest systems.
“Yes, yes, we know Fujitsu and Sun have the entire market in their hands and no one else is selling anything – can I laugh now, or do I have to wait a bit longer?”
Never said they did, but what I think matters, is showing evidence of CPU/HW/stack R&D, delivering capabilities and performance that customers want, at competitive price points and being consistent over time. Proven history and delivery matters. and TCO of course.
“Oh, and please: do NOT confuse Oracle’s bottom line as a company and its hardware division’s bottom line. Again, abyssal difference.”
The bottom line is how well the company is doing as a “whole” cause that’s what investors care most about. Oracle is now a technology company selling software and hardware that are engineered to work best together while maintaining open-ness and offering choice. Oracle is showing significant sales growth in its engineered systems as well as mid to high end SPARC systems as reported in last fiscal quarter. Oracle SW drives HW sales and vice versa so theres a lot of co-dependencies lately. Remember that Oracle acquired Sun just 3 years ago and takes time to turn around a big ship.
“The Oracle hardware business continues to hitch a ride at the expense of mine and other’s software maintenance revenues. Take that away and the entire Sun business would have been ejected years ago as the loss-leading and loss-making piece it truly is. Careful, it’s not just me who can read a company’s annual prospectus and analyze it.”
Don’t agree about maintenance revenue but its impossible to prove one way or the other. Many of Oracles SW products have had boosts since the introduction of Engineered Systems and now Solaris based systems, mostly coming from SPARC. And if you look at latest FY13 report, 74% of Oracle Revenue is SW. http://www.oracle.com/us/corporate/investor-relations/financials/q4fy13-1961623.pdf ,but it shows 28% of Revenue is “New software licenses and cloud software subscriptions”, 14% of revenue is “Hardware Systems Revenues”, 46% “Software license updates and product support”. No breakout of maintenance but clearly lots of revenue from new licenses.
“No one (outside of Oracle) in his right mind would use 12c for anything near production for the next 3 years, given the pitiful history of Oracle software stability in the past 20 years.”
We shall see, Beta begins soon…All the analysts are raving about it. Why dont you become a beta tester so you can air out the laundry?
“As for my liking or not of Oracle – the software, NOT the hardware! – I’ll leave you with a couple of thoughts:
I’ve been using it for 25 years, first at Oracle, then outside and I have yet to find a better piece of database software, once it’s been stabilised – that is why I keep using it and recommending it.”
Great! Well that’s the whole point about the “new” Oracle. Your challenges with “stabilization” is probably from the huge complexities of having 5-10 vendors products to deal with across the SW/HW stack. Its not just the DB running, but OS, middleware, virtualization, server, storage, networking etc. And who knows if it was actually the DB that wasn’t stable, maybe some other layer in stack? But that’s Oracles focus today- eliminating those inter-layer complexities to provide highest stability as well as reduced management and clearly costs. It’s the first time in the industry that the DB engineers are on same team as engineers in rest of the “stack”-all working on improvements.
And so you appear to agree that Oracle DB is still competitive and its legacy remains strong. Its now Oracles opportunities to grow the business around the DB.. And figure out how to get you and others as an end to end customer.
“The place where I work would have moved to M$SQL years ago if it wasn’t for my efforts to keep Oracle here – we shove over a quarter of a million bucks up Larry’s bottom line as a result.”
Oh god forbid ;-)! Otherwise you would have filled M$ coffers instead. Atleast Oracle has a cool sailing team 😉 Want to be CEO? I hear theres an opening. 😉
God Speed!
Oh, we more than agree on what Oracle is doing and how.
I just disagree that I have to pay for it. 🙂
Still, I suppose technology companies live off R&D, and of course a good portion of that comes off maintenance revenue. As such, it’s only fair.
I’m always reminded of how Prime Computers met its end: an ex-IBM “expert IT genius” CEO walked in, looked at all the expenses, saw this huge expense column in a spreadsheet and decided to cut it to nothing.
Three years later, Prime went into receivership.
The column? R&D…
Some “expert”! I wonder if he wasn’t a plant…
Being a beta tester for 12c? Never! I refuse to be such for as long as I don’t get a maintenance discount on my existing licenses.
Beta testing – done properly rather than as a “wow, that’s kewl” exercise – requires a lot of time and resources and I’m not paid by Oracle nor do I work for them anymore, so no go.
That report is very enlightening for those willing to spend the time to study it carefully. Worth the investment in time, folks.
It’ll be very hard for Oracle to get a lot of customers like us on the end-to-end bandwagon.
We run a very large mix of software and hardware. Intentionally.
Basically, the hw platform for us is irrelevant provided it can run ALL the software. We use a mix of M$, IBM, Oracle, GNU and FSF for software.
Last time I looked Exa* from Oracle only runs Oracle stack with the notable exception of Hadoop (hinthint). As such it’s a no go for us and will remain so for a while.
It’s called general purpose computing for a reason. As opposed to lock-in – which is what Oracle is trying. IBM was there 50 years ago and got burned, so good luck to Oracle on that!
As for CEO: hmmm, pass. I like my kids to know who I am!… 🙂
Hey-Thanks for the agreement- I’ll take that as a compliment.
Yeah, Technology companies *depend* on R&D to advance, and without revenue, but more importantly, margin and profit, its not possible, so its Oracles success as a company that’s feeding back into R&D, which I believe ultimately benefits the customer with better, more competitive products. And yes, Beta testing isnt for everyone, but you can always negotiate 😉
Did you see this new article? It highlights how Oracle Hardware is driving the bottom line: http://bit.ly/1bGDAMT . This quote says it best “Considering that hardware is a small part of Oracle’s overall business and that its most important business–software updates–has shown strongly improving profitability, it is easy to see why Ellison has called the Sun acquisition the best acquisition he has ever made”
By the way, you keep talking about maintenance costs. Have you ever looked at IBM’s or HP’s service/maintenance costs, and if any of them are fixed on yearly basis and what you get? Especially look at “the after 3 years” costs and you’ll see what I mean. Oracle’s fixed 12% Premiere Support is all inclusive and fixed year after year after year. No hidden costs, no surprises/gotchas after 3 years, you know what you expect to pay, etc. Last time I compared to offerings from IBM, I calculated on average a 30-50% SW/HW maintenance/support cost relative to Hardware cost.
But I’m not sure if I agree in your hardware platform is irrelevant comment, as clearly, if you were running a very slow system, although it may run all your software, wont produce the results you need fast enough. Time is money after all. Part of that time is in what the system delivers (throughput, response times, scalability, etc), and part of that time is how much time it took you to get system up and running (integration/installation/testing/validation/etc) and stay running (uptime, availability, lifecycle, upgrades/patches/etc), in production of course.
And by the way, Exa* is not all of Oracles engineered systems, as they are purpose built engineered systems for specific Oracle SW (Exadata vs DB, Exalogic for middleware, etc) so yes, they may interact with 3rd party SW but clearly not all cases. Oracles “SuperCluster” engineered systems *are* general purpose and run the 10,000+ apps certified/qualified on Solaris-practically any software needed for an enterprise. And then of course, Oracle is still building best of breed servers based on latest Xeon and SPARC running either Oracle Linux or Solaris. So there is choice in Oracles “Hardware” to run *all/any* of your software needs.
And no, theres no chance of lock-in with Oracle as you can easily switch/have choice on any layer of the stack. IBM’s problem is that many of their technologies *required* you to go to IBM, especially their IGS, which is where +50% of their revenue comes from. AIX requires Power, anything Mainframe requires IBM, Power is only available from IBM, etc. Blades, I believe is the biggest form of lock-in in the industry.
With Oracle, theres no requirement of any product to buy only from Oracle, including Oracles Engineered systems (except where its specialized to only run certain Oracle SW of course), but theres no special programming/SW development needed to run and you can even run 3rd Party apps like SAP and so on. But clearly, Oracles focus is on “persuading” customers to run Oracle on Oracle by proving lower TCO, faster time to deployment, lower complexities, reduced admin/management, etc. But the choice is yours.
@Phil_Dunn I encourage you to write these essays on your own blog and link to them in comments. I get a crazy amount of traffic from search engines matching the terms you use in your lengthy, prosaic, interactions with my blog visitors.
Phil
Last time I checked it was bad practice for vendors to accuse customers of lacking education or failing to do their homework. There are two ways that can play out:
1/ The customer didn’t do their homework and hasn’t been educated on your product. As the vendor there is only one person to blame here – and that’s you. Vendors don’t have a right to expect customers to know and love their products.
2/ The customer, as in the case of Noons, is very well educated, has done lots of homework (more than you, it appears) and simply doesn’t like your product, doesn’t believe the marketing claims and has encountered another vendor who made a better case. Again there is no point looking at anyone else but yourself.
I’m struggling to think of a scenario in which telling the customer they know “NADA” is going to come out well for you. Surely it’s a no-win strategy, right?
Oracle has a certain reputation for aggressive behaviour towards customers which I think is unfairly applied by some to all its staff. I know many Oracle employees who genuinely care about their customers and treat them with respect at all times. Right now it’s not looking as if you are one of them.
I assume you do really work for Oracle, right? You’re not someone working for a competitor and trying to make Oracle look bad?