mprosrv client count

[bdowne01]

Hi Everyone,

I'm a unix admin that helps manage a large character-based progress application (3000-3500 users). The application is structured in a way that the users connect to a front-end farm of servers (RedHat Linux), and use a remote connection back to a larger database server (AIX).

Our DBAs have configured the database to have a _mprosrv process for every 10 remote connections. The end result is the AIX server experiences extremely high kernel CPU% time, higher than user %--the end result being the server spins its wheels most of the time. IBM stated the kernel time is being generated by the large amount of _mprosrv processes.

So my question is what is the "best practice" for setting this up? Is there one? What should be a decent amount of _mprosrv processes to handle 3000 remote connections?

Thanks everyone!

 

[TomBascom]

That's a rather cheeky bit of stating the obvious on IBM's part

5 to 1 is more typical. But given the paucity of information about your system I can't say if it is appropriate to you situation.

With that sort of workload it could be an awful lot of things. I would guess that you probably have a few "tuning opportunities" available.

But you'll need to post a whole lot more information about your system, your release of Progress, the workload and performance metrics to get anything better than vague & fuzzy responses.

Or you could save a lot of time and hire a good consultant to come in and take a look

 

[TomBascom]

Just guessing but...

Given the user count and the problem description I'll go out on a limb and speculate that you have a very high "logical io" rate. If you were to look in PROMON I'd expect that you will find that it is mostly read activity and that you are doing hundreds of thousands (or millions if you have a sufficiently beefy server) of reads per second.

It would be fascinating to look deeper and see which tables and indexes are the most active. I'll go out on another limb and bet you a beer that you (or your developers) would find some surprises there.

You may, or may not, have much going on in terms of actual disk IO. If so that suggests one line of inquiry.

You might, or might not, have many "latch timeouts". If that is so then another line of inquiry opens up.

It would also be interesting to know how much traffic netstat says there is on your network.