General
cancel
Showing results for 
Search instead for 
Did you mean: 

Performance problems, Progress database on HP UX 11.0

SOLVED
Go to solution
Jay Newman
Frequent Advisor

Performance problems, Progress database on HP UX 11.0

Hi there.
I have a question for any admins who work with systems running Progress databases.
We recently moved from HP UX 10.20 to 11.0, and ever since we have seen "spikes" where a user (or several users) suddenly are frozen for up to a minute. Usually it resumes where it left off, but occasionally the user gets bumped off.
At the same time, we moved to a much faster server (a K460 instead of a G50), with more CPU's, bigger disks, etc..
I am starting the whole process of install Glance, etc., but since this application is new to this server, I don't have a healthy baseline to compare against.
Has anyone here encountered database performance issues on upgrading to HP UX 11.0 ?
"Success is defined by getting up one more time than you fall down."
8 REPLIES
Sridhar Bhaskarla
Honored Contributor

Re: Performance problems, Progress database on HP UX 11.0

Jay,

I didn't work with the systems that had Ingres database. YThe spikes could be due to several factors. As I understand that you are configuring Glace+, check this thread to see if it can help you out.

I would suggest you to configure Ingres as an application (workload) in SCOPE/UX and analyze
it in detail.

http://forums.itrc.hp.com/cm/QuestionAnswer/1,1150,0x46e7d211e18ad5118ff10090279cd0f9,00.html

-Sri
You may be disappointed if you fail, but you are doomed if you don't try
Paula J Frazer-Campbell
Honored Contributor

Re: Performance problems, Progress database on HP UX 11.0

Hi Jay

Faster server more cpu and new os, but performance problems all points to kernal config.

Periodic freezes point to buffer cache.

The attched doc should help.

Paula
If you can spell SysAdmin then you is one - anon
Jay Newman
Frequent Advisor

Re: Performance problems, Progress database on HP UX 11.0

Thanks; that link may help me when I get to the point of deep analysis of the system.. but without a baseline of the system working correctly, I could be looking for a needle in a haystack.
(By the way, it is a Progress database, not Ingres)
If we are lucky, some other admin has made a similar O/S transition and can let us know if there were any issues such as kernel parameters or patches etc.
"Success is defined by getting up one more time than you fall down."
Joe Gilligan
Occasional Visitor

Re: Performance problems, Progress database on HP UX 11.0

Well, I've had alot of issues that I've resolved regarding the database performance. I would suggest looking at how you are utilizing the database. As mentioned before kernel tuning is essential. I found that the dbc_min_pct values should not be allowed to float very much. Your swapchucks value may also need to be increased if you are suffering from a disk or i/o bottleneck this is a function of memory management. Progress isn't too much of hog on memory but requires about 2Mb/user and needs to be checked.

Hope this helps - as late as it is.
Hey, what's this do?
Shar Hunter
Frequent Advisor
Solution

Re: Performance problems, Progress database on HP UX 11.0

We run Trend, a Progress application, on HPUX 11.0

We were on 10.2

The Progress is 9.1 I think.

We have not had any performance problems, other then an ongoing printing head ache.

Printer jobs randomly do not finish, about one a week. But I am working on that.

There is a lot you can in the area of performance tuning on Progress.

But I am mostly a novice, so only if you are interested will send some articles on how to tune Progress databases.
(they are to big to post here)

You can try:
www.progress.com/support/techpapers/tuning.htm

If you do not have Progress tuned well, it will run poorly and even behave strangely. Ha ha, I know that from experience.


What application are you running?

Tro
I don't think I'm in Kansas anymore.
Bill Hassell
Honored Contributor

Re: Performance problems, Progress database on HP UX 11.0

11.0 has a history of strange response delays. Most of them have been fixed with patches, so before going very far, make sure you have all the SupportPlus patches installed from either June or September. As always, check the patch recommendations to deselect or replace any patches in the bundle.

Note that there is no one cause so looking for specific patches that address the symptom may not have any effect in your situation.

Something else to look at is sequential access, something that always occurs during a backup (we're talking about files, not raw volumes), or during serial database access. Changes made to buffercache queueing in 10.20 and more so in 11.0 give preference to adjacent disk activity, which can sometimes lockout random tasks like bdf or login or vi for seconds to minutes.

To help with this symptom, a new kernel parameter called disksort_seconds was introduced to helpp reduce the delays. The tunable parameter can be set to 0, 1, 2, 4, 8, 16, 32, 64, 128 or 256 second(s). If "disksort_seconds" is 0 (default value), the time stamp is disabled, which means that time aspect is not taking effect.

Now picking a value is tricky...a low value such as 1,2 or 4 will tend to defeat the performance advantage of the buffercache's preference for sequential I/O. I would start with 8 and then armed with Glance, take a look at the disk queues. You won't need a baseline with Glance since users will tell you when the system is too slow. Use the 'o' screen to limit visible processes to interesting ones, perhaps 5 secs of CPU and 5 I/O's per second.


Bill Hassell, sysadmin
Joanna Clarke
Occasional Visitor

Re: Performance problems, Progress database on HP UX 11.0

Hi,

We have recently upgraded from 10.20 to 11.0 32-bit and are using Progress 8.3e. Unfortunately since the upgrade we are having the same freezing problem as Jay. I was just wondering Jay if you could perhaps let me know if you found a solution.

Alsowe are currently trying Bills solution. Last weeks we changed disksort_seconds to 8 but this didn't help so last night we moved it to 16.
Please help as no-one at HP or Progress seem to know a solution.
Thanks
Jay Newman
Frequent Advisor

Re: Performance problems, Progress database on HP UX 11.0

Thanks for the responses.
As is, it appears the main problem was that, as we added more application users, we consumed all the available read buffers allocated on the Progress database.
(The default is 150.)
We brought the number of read buffer up to 8K, and immediately saw a marked improvement in performance.

Here is the change that was required:
Old startup command for the database :
/opt/dlc/bin/proserve -L 4500 -n 80

New startup command for the database :
/opt/dlc/bin/proserve -B 8000 -L 4500 -n 80

The suggestions regarding kernel tuning were appreciated, but we found that the values already had been set to appropriate numbers.

Running Glance and sar was inconclusive, as the problem was intermittent and generally did not last long enough to gather statistics.

"Success is defined by getting up one more time than you fall down."