Dave Moore discusses the relatively new set of OS performance data collections scripts on his webpage here. Dave writes:
OS Watcher is a utility provided on MetaLink (Note 301137.1) primarily for support issues in a RAC environment. I must admit I was captivated by the name and wondered if I could use this tool instead of expensive 3rd party products for monitoring key operating system metrics. The verdict is “no” and I was less than impressed.
OS Watcher is a series of shell scripts that run on AIX, Solaris, HP-UX and Linux. Simple commands such as ps, top, vmstat, netstat and others are executed at regular intervals and their output is appended to a file in a directory specific to that command.
I have not personally taken the time to play with these scripts (I have my own), but I can read the tea leaves. Oracle support will most likely start asking for this data for any problem you might be having (regardless of whether you have a performance related problem). It might be smart to start collecting this data so you don’t hear something like, “Please reproduce the problem after installing OSW.” Just a thought.
I’ll see if I can arrange a test of how heavy the collection of this data is and blog on what I find. I read through the scripts and it looks like some pretty heavy collection. I never liked performance monitoring tools that carry a heavy “tare weight”. Did any of you use CA Unicenter in the old days?
If you have a Metalink account the toolkit is available here.
Kevin Closson's Blog: Platforms, Databases and Storage 
OS Watcher is no more “heavy” than the underlying utilities it executes. Since it is user-configurable, it is a trivial matter to adjust collection intervals to suit individual needs. The collection interval and archival timeframe are both settable by command line parameters. We find the default values of 30 second snaps and 48 hours archive are quite acceptable to most users.
OS Watcher was created to address the problem of lack of operating system performance context when analyzing performance issues reported for the database. And, while Statspack has been historically quite valuable for analyzing performance issues, a logical and effective performance methodology must also take the operating system into consideration.
As is often the case with performance issues, DBAs as well as Oracle Support engineers typically find themselves responding to performance incidents through a rear-view mirror. We may be told this afternoon there was a horrible performance incident this morning. And, we are expected to offer expert analysis hours after the incident is over. Many will recall the days of utlbstat/estat, where performance data had to be collected realtime and responding effectively to such crises was a nightmare. The advent of Statspack, with its perfstat schema archive addressed that problem beautifully.
Having a Statspack (or now, AWR) archive available for after-the-fact reporting of performance problems is critical. OS Watcher similarly adds archival of operating system peformance data to the bag and completes the analytical picture for many issues.
We have found that, while many customers do collect operating system data through various tools and methods, the collection is often inconsistent and equally often produced in data formats not readily usable to support engineers. Thus OS Watcher was created. The design specifically considered deployment issues across platforms, and the decision to rely on common Unix/Linux utilities such as vmstat and iostat was a good one, I believe. It is shell script based and is quite portable.
Since OS Watcher is not a particularly “heavy” tool to install and run, we are finding customer acceptance is quite positive and widespread. The data collection archives are customer-controllable and at the same time very useful for end users as well as Oracle Support engineers.
One reason RDA was so well accepted is that the data content and output format– html for RDA– are directly usable and beneficial to our customers. OS Watcher, with its text output format, achieves that same benefit and we hope its acceptance continues to grow.
“Did any of you use CA Unicenter in the old days?”
In the old days? Some of us (or them if you prefer) are still using this monster.
Alex,
I haven’t seen CA Unicenter in action for quite some time now. You mean it hasn’t gotten any better?
Seems like the idea is taken from the “VOS Watcher” utility which runs similar set of OS commands including some additional.
Kevin, how can I put it… I don’t see any way for those tools to get any better except if someone will rewrite them completely.
One of our clients uses Unicenter – I’m not very deeply (thanks God) involved into it but what I use it for is enough to see it in my nightmares. Architecture of a dinosaur and usability on level -2 if you ask me evaluate it on the scale from 0 to 5.
On another site I was a “lucky” user of Control-M from BMC as well as Patrol. Control-M does look a little bit better (IMHO) compare to Unicenter Workload but I would still never choose it if asked (well, that site is historically a mainframe shop).
Back to the rewriting from scratch – there are lots of open source tools that have good ideas and framework but far from production quality (I mean real enterprise quality). My understanding is that commercial software development in this area is problematic because potential consumers are divided in two major camps:
1. Why would we pay for a tool if there are zillion free open-source products?
2. We can’t trust those freshly-baked potatoes and better use rock solid “proven” products. Here might fall existing users as well.
Anyway, I’m probably a greenhorn in this area anyway so you better don’t listen to me. 😉