Network Appliance OnTap GX for Oracle? Clustered Name Space. A Good User Group (OSWOUG)

Presenting NAS Architecture at OSWOUG
I had a speaking session recently at OSWOUG about NAS architectures as pertaining to Oracle. It was a good group, and I enjoyed the opportunity to speak there. The presentation was an “animated” digest of this paper and covers:

  • single-headed filers
  • clustered single-headed filers
  • asymmetrical multi-headed NAS
  • symmetrical multi-headed NAS

Symmetric NAS versus Clustered Name Space
Since the only model in existence for the symmetrical multi-headed NAS architecture is the HP Enterprise File Services Clustered Gateway (which is an OEMed version of the PolyServe File Serving Utility), I spent some time discussing the underpinnings—the cluster filesystem. At one point I spent just a few moments discussing Cluster Name Space technology, but I wish I had taken it further. I’ll make up for it now.

If You Can’t Cluster Correctly, At Least Cluster Something
Something is better than nothing, right? Many years ago, an old friend of mine was coming back from duty in Antarctica on a coast guard ice cutter. You’d have to know Kenny to get the full picture but at one point, over beer, he pulled out a phrase that still makes me chuckle. I don’t remember the context, but his retort to some rant I was on was, “Well, if you can’t tie a knot, tie a lot”. That, my friends, is the best way to sum up Cluster Name Space NAS technology. I couldn’t find a dictionary definition, so I’ll strike a claim:

clus·ter name space'klus-ter 'nAm 'spAs

Software technology which when applied to separate storage devices gives the illusion of scalable filesystem presentation. Cluster name space technology is a filesystem directory aggregator. Files stored in a directory in a cluster name space implementation cannot span the underlying storage devices. Single large files cannot benefit from cluster name space technology. Cluster name space is commonly referred to by the acronym CNS. When applied to Oracle databases, CNS forces the physical partitioning of data across the multiple, aggregated underlying storage devices.


To apply a virtualized presentation layer to a bottlenecked storage architecture: “hey, let’s just CNS it and ship it”

The verb form of the term was demonstrated when Network Appliance bought Spinnaker back in 2003. When you spend 300 million dollars on technology, what is done with it often winds up looking more like a verb than a noun. What do I mean by that? There is nothing about Network Appliance’s OnTap GX technology that was built from the ground up to be clustered and scalable. Network Appliance has had clustered filers for a long time. That doesn’t mean the data in the filers is clustered. It is merely a cabling exercise. What they did between the 2003 purchase of Spinnaker and the release of OnTap GX was a verb—they CNSed their existing technology. I can tell from the emails I get that there are about 42 of you who think I make all this stuff up. Let’s get it from the horse’s mouth, quoting from page 8 of this NetApp whitepaper covering OnTap GX:

When an NFS request comes to a Data ONTAP GX node, if the requested data is local to the node that receives the request [reference removed], that node serves the file. If the data is not local [reference removed], the node routes the request over the private switched cluster network to the node where the data resides. That node then forwards the requested data back to the requesting node to be forwarded to the client.

Perhaps a picture can replace a thousand words. Nobody, not even in the deepest, darkest, remote areas of the Amazon rain forest would mistake what is depicted in the following graphic for a symmetrical file serving technology:


The Press
There has been a lot of press coverage of Data OnTap GX (NetApp is the poster child for CNS), and most articles get the technology just flat wrong. This article, on the other hand gets it quite right:

NetApp’s SpinServer is a clustered file system for NFS that currently doesn’t support CIFS. SpinServer exports a single namespace across all storage nodes, but files aren’t striped across nodes

And so does this article which exposes the weakness of the CNS verb situation:

…immediately found a problem with the high-availability (HA) failover detection system. A failure could be bad enough that clients could not access data but not bad enough to alert the system.

What Does This Have To Do With Oracle?
OK, this is a blog about Oracle and storage. Let me interpret the quote from NetApp’s paper covering Data OnTap GX. If you have a large Oracle datafile that gets hammered with I/O this technology will do absolutely nothing to improve your throughput or reduce I/O latencies. A single file is contained within one filer’s storage and thus, represents an I/O throughput bottleneck. One of the sites I read, DrunkenData, has a thread on this topic. It appears as though the thread is aimed at trying to believe in something more than to understand something, but my assessment could be wrong.

The CNS topic is quite simple; the word “cluster” in the CNS context has nothing to do with file-level scalability. Yes, I know Network Appliance is a multi-billion dollar corporation so I’m sure my exposé of this topic will be met with nefarious men wearing trench coats lying in wait to attack…

Cram Some Technology into Your Solution
This article covers some clustered NAS topics, but there was one bit in there that stood out when I read it and I’d like to comment. The following excerpt recommends CNS technology for Oracle:

…To avoid this problem, consider a global namespace product…The only way to implement a cross-platform global namespace is to replace your NAS infrastructure with, for example, NetApp’s SpinServer or Panasas’ ActiveScale. If Oracle 10 on a Linux cluster is in your future, then the NetApp and Panasas solutions should be on your short list.

Bad mojo! I have said many times before that Oracle on NAS is a good model and NetApp is obviously the 800lb gorilla in this space. However, I disagree with the idea posted in this quoted article for two reasons. First, Oracle uses large files and CNS does not scale single large files. To get the benefit of CNS, you will have to partition your large files into smaller ones so as to get multiple filers serving the data. If you wanted to physically partition your data, you would have chosen a different database technology like DB2 EEE. And secondly, Oracle has an established program for NAS vendors to certify their capabilities with Oracle databases. Panasas may be good technology, but since it is not an OSCP vendor, it doesn’t get to play with Oracle. On the other hand, the HP EFS-CG, which is multi-headed and scalable, is. Fair is Fair.

A Change of Heart
Here was an interesting post in the mailing list about “global name space”. I don’t know how the different filesystem technologies got clumped together as they did, but that was two years ago. I found it interesting that the poster was from Verari Systems who now resell the PolyServe File Serving Utility.

I hope you know more now about CNS than you would ever care to know–literally.

5 Responses to “Network Appliance OnTap GX for Oracle? Clustered Name Space. A Good User Group (OSWOUG)”

  1. 1 Carl Johnson November 15, 2006 at 6:38 pm

    Does the “Striped Volume Sets (FlexVol High-Performance Option)” described on pages 17-18 change your thoughts on this?

  2. 2 joe noname November 27, 2006 at 12:17 pm

    the striped-volume-option is just a joke:

    you still can only NFS-mount one specific NFS-filesystem from ONE head.
    (NFS simply has no “multi-server-mount” – ha ha )
    So each write()/read() from one NFS-client – will always only use exactly ONE head for read()/write().
    What this server-head does with the data given to it – this is a different story – but whatever it is – it’s for sure adding more junk into the datapath. if that’s what scalability means to you – cool.

  3. 3 steve lambourne December 7, 2006 at 5:47 pm

    are your thoughts ths same regarding BlueArc CNS

  4. 4 kevinclosson December 7, 2006 at 7:03 pm

    I have to admit that I haven’t dug into BluArc sufficiently. If it is nothing more than a CNS, than I will still point out that making multiple NAS heads “feel” symmetrical is just not going to cut it.

    Thanks for the comment.

  5. 5 kevinclosson December 8, 2006 at 5:25 pm

    OK, I am not surprised that BlueArc CNS is just CNS. I think BlueArc is some of the most innovative single-headed NAS out there, but putting “glue” above the NAS heads to make them “feel” like one is not scalable. You still get “silos” of data. That is, all I/O for any given file must be serviced by one, and only one, NAS head.

    Architectural trade-offs.

Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.


I work for Amazon Web Services. The opinions I share in this blog are my own. I'm *not* communicating as a spokesperson for Amazon. In other words, I work at Amazon, but this is my own opinion.

Enter your email address to follow this blog and receive notifications of new posts by email.

Join 746 other subscribers
Oracle ACE Program Status

Click It

website metrics

Fond Memories


All content is © Kevin Closson and "Kevin Closson's Blog: Platforms, Databases, and Storage", 2006-2015. Unauthorized use and/or duplication of this material without express and written permission from this blog’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Kevin Closson and Kevin Closson's Blog: Platforms, Databases, and Storage with appropriate and specific direction to the original content.

%d bloggers like this: