system tuning and bottle necks

[vmsserbo]

How do I determine if my systems need tuning or have bottlenecks. I work with about 80 vax systems running openvms. Are there commands I can use or reports ?

[Arch_Muthiah · Accepted solution]

Miles,

The best utility is DECset, under this there is a DIGITAL
Performance and Coverage Analyzer (PCA, which can be used to collect and analyze various types of system data.

http://h71000.www7.hp.com/DOC/decset.html
http://h71000.www7.hp.com/DOC/73FINAL/5605/5605_pro.pdf


Archunan

[Arch_Muthiah]

Miles,

The utility you can use is DECset, under this there is a DIGITAL
Performance and Coverage Analyzer (PCA, which can be used to collect and analyze various types of system data.

http://h71000.www7.hp.com/DOC/decset.html
http://h71000.www7.hp.com/DOC/73FINAL/5605/5605_pro.pdf


Archunan

[Karl Rohwedder]

If your users are satisfied, you don't need tuning :-)

Tools you can use:
- MONITOR (included in VMS) to view various aspects of the system live, offline or via summary reports (see SYS$EXAMPLES for a DCL routine to setup a regular job)

- T4 (you can find it via the OpenVMS homepage) can create very nice graphic reports and allows to store history data

- TDC (The Data Collector) together with ECP can also create nice graphical or textual reports

- there are some more comercial products

- not to forget AMDS or AVAILMAN, which give are nice online overview (green/yellow/red) od CPU,MEMORY,IO

We use T4 and ECP to automatically create reports and store historical data.

regards Kalle

[Arch_Muthiah]

Miles,

We have MONITOR, ACCOUNTING, and AUTOGEN utility also, using which we can understand and improve the performance of the three major hardware resources---CPU, memory, and disk I/O.

AUTOGEN mostly will do best tuning in terms of System Parameters.


Archunan

[Duncan Morris]

Miles,

the best reporting tool will be your users!

You can always tune a system to get a little more mileage out of it, but is the effort of tuning rewarded by the gains? If your users have specific areas of concern, then you should check out these areas.

You can find performance and tuning guides in the VMS documentations set. See OpenVMS Performance Management here

http://h71000.www7.hp.com/doc/os73_index.html

[Phillip Thayer]

Miles,

You should always listen to what your users are saying as far as system performance. However, you still need to have some kind of proactive monitoring in place to be able to spot problems before the users start flooding your phone with calls that the system is slow.

MONITOR, T4, ECP, DECSet PCA, AMDS and Availman are all tools that can provide these type of proactive reports to possibly see problems before they occur. Check the performance tuning manual referenced with a link in a previous post for what you should be looking for.

The thing to remember also is that you can overtune th system and actually make it run worse than it is now. If you don't see anything on the reports that points to a problem based on the Performance Tuning manual then don't touch the system and let it go on chugging. Also, with 80 system to manage you should probably look to something that can provide a summary screen with the capability to drill down into systems that are flagged with problems. That would mean AMDS would probably be your best bet for keeping track of all the systems.

Phil

[Arch_Muthiah]

Miles,

You may be knwon this manual, "OpenVMS Performance management", It is very good manual to understand OpenVMS system's performance related things.
You will become so confident on OpenVMS once complete reading this manual.
http://h71000.www7.hp.com/DOC/73final/documentation/pdf/OVMS_73_perf_mang.PDF

This manual addresses system managers and other experienced users responsible
for maintaining a consistently high level of system performance, for diagnosing
problems on a routine basis, and for taking appropriate remedial action.

OpenVMS System Dump Analyzer Utility Manual also helpfull system manager who must investigate the causes of system failures
and debug kernel-mode code, such as a device driver level.

Thanks
Archunan

[Ian Miller.]

Computer systems are bought to run applications therefor it is application performance as percieved by the users. Of course user perception is subjective but unfortunally they are the people paying the bills.

You need to determine the key performance indicators and measure those. These are often the time to perform some important operation or the number of some operations performed within a certain time.

Also note the important time is not measured on the vms system but it is the time as seen from the users terminal.

The tools mentioned are excellent for measuring system metrics but system metrics like I/O per second are not what you should concentrate on. The tools are good for investigating performance problems.

The VMS Performance Management manual
http://h71000.www7.hp.com/doc/73final/6491/6491pro.html
or in pdf
http://h71000.www7.hp.com/doc/73final/documentation/pdf/OVMS_73_perf_mang.PDF

is the one to start with. The tools already mentioned like T4 and EPC have manuals which are good for understanding the tool but not for understanding your overall performance management.

The PCA tool (part of DECset) previously mentioned is a tool for looking for performance problems in a specific program. This would be useful when you have determined that a specific application is not performing well enough.

What aspect of system performace are your users complaining about?

[Arch_Muthiah]

Miles,

Is this for VAX openVMS?

Then you should use "Dynamic Load Balancer Plus", this is specially for VAX/OpenVMS system.

Dynamic Load Balancer Plus (DLB Plus) is the original OpenVMS on-line dynamic tuning product. DLB Plus monitors the system load, any bottlenecks and makes adjustments to process memory usage and CPU scheduling parameters. These adjustments enhance system performance.

DLB Plus makes adjustments dynamically, continuously as system loads change. Dynamic tuning results in a balanced and well-tuned OpenVMS system.

Benefits include:
efficient use of CPU and memory resources
more consistent response times continuous adjustment based on the current environment conditions.

Archunan

[vmsserbo]

I am using Vaxes . Give me examples of commands using Dynamic Load Balancer Plus

MIles

[Arch_Muthiah]

Miles,

You can download DLB Plus here....
http://www.ttinet.com/dlbplus.html

And you can find the docs here...
http://www.ttinet.com/doc/dlbplus_clean.html

hope this solve your problem.


Archunan

[Arch_Muthiah]

Miles,

Once you install DLB Plus, Just start DLBPLUS using DLBPLUS_STARTUP.COM
and it is a menu based and start main menu using DLBPLUS.COM, there you can find so many options you can choose....



Archunan

[Guillou_2]

Hi,

you can find information about performance and bottleneck at this link:
http://h18000.www1.hp.com/support/asktima/operating_systems/009BD243-C877FC20-1C0186.html

regards

Steph.

[comarow]

Do you have a support contract? We have a document called the performance cookbook which is a step by step method for looking at a system.

The first step has to be Monitor.

Monitor system will look at CPU utilization,
CPU queue, memory and overall activity.
The CPU queue is especially useful as if you have it stacking up on an AVERAGE of 1 of more, you have a CPU bottleneck.

$monitor DISK/item = queue
OTHER than when backups are running
(backup purposely queue up I/Os)
If you see on the AVERAGE queue depths of 1 or greater, the disk is saturated. Often you can simply move hot files off the disks with a problem.

$monitor page
will look at your page faults. However, that takes some understanding to look at.
The Performance Cookbook provides much information.

You can create monitor reports. In
Sys$examples there are several automated monitor programs which will monitor your system and send you the reports. I personally prefer 15 minute snapshots when your system is busy.

$show system
itself is a very useful command. It will show the process states. Are there any unusual states? A lot of Com stated.

$mon lock
will show the number of locks, and their states. Are there a lot waiting?

$show mem/file
will show how much space is left in your pagefile. If your pagefile gets low in free space, it will crawl, and at around 0 free your system will hand.

$show mem/full/pool
compare the initial size of npagedyn to the current size. If it has significantly expanded time to run autogen with feedback.
You might wish to put the current size as the min size
min_npagedyn=current
when you run autogen with feedback
$autogen savparams setparams
Your system must have been up for at least 24 hours and make sure it had the full load.
Look at the report and make sure there were no fatal errors.


Similarly, when you do $show mem/full/pool you want PAGEDYN to be at least 50% free.

Someone suggested dynamic load balancer. Sometimes that may work, but I've seen systems where IT was the problem. We turned it off and improved performance. I tend to think it was better in earlier versions of VMS.

VIOC is a cache to improve I/O. In more recent versions of VMS.
$show mem/cache/full
will show the size and how well it is working.
We have an article on reading it and using it.

VCC_MAXSIZE is the sysgen parameter that will increase it's size.
add
MIN_VCCMAXSIZE to modparams.dat

Of course, the ultimate test is to log on to the system, try a number of command, including the applications. You certainly can have application slowdowns.

Finally, I would like to point out that
Colorado Support does offer VAST or custom consulting, and can analyze your performanc situation, identify bottlenecks and offer solutions.

[Robert Gezelter]

Miles,

While listening to your users is always a good idea, it can be quite misleading.

I have encountered many sites where truly miserable performance is accepted as normal, because it has been that way for a LONG LONG time.

While all of the sophisticated tools are nice, you can glean a huge amount of information from simply running MONITOR and looking at several parameters.

Good tuning is a question of identifying what is limiting performance, and fixing that issue. Then iterate again, fixing the next problem. When you have a system where there are no factors that are limiting performance (other than shortage of work), you are done, at least for the moment.

- Bob Gezelter, http://www.rlgsc.com

[Allan Bowman]

The simplest solution is to use MONITOR and AUTOGEN on a regular basis.

You can set up MONITOR as a continuously running process with daily reports, or you may just need to run it manually for a few minutes once a week - it totally depends on how volatile your environment is. If you are not adding new applications or lots of users, you may just need to check things out once a month or so. If there are frequent changes, you should run it often, or use one of the other tools that others have mentioned.

You can periodically run AUTOGEN with FEEDBACK just up to the reporting phase. Look at the reports generated and see if any of the recommended changes appear to be significant enough to implement.

If your 80 systems are identical, you should only need to look at one of them on a regular basis and apply any changes to all of them.

Allan in Atlanta

[Jan van den Ende]

Miles,

although I do agree with Allan that 80 more-or-less identical systems should reduce your effort considerably, in my view that is for a totally different reason.

Collect your metrics (any of the aforementioned tools), but generally you have only to look at the summaries, and COMPARE THOSE.

Any system sticking out is the one that deserves special attention.

If a system appears to do much worse than the others, you need to find the cause. If it is just a lot of extra load, maybe that can be re-distributed? Anyway, the heaviest loaded system will be the one that gives the strongest response to any changes, so that is the one you will need to work on.

If a system does much BETTER than the others (without just having little to do) investigate THAT one also! It might just tell you how to improve all others!

hth

Proost.

Have one on me.

jpe