K370 and Sybase ASE

[Mike Hassell]

I have a K370 with four PA-8200 processors installed, running HP-UX 10.20 and ASE 11.9.2. We have on average 100 or less users connected to the database at one time. This is a custom written database and is the only application running on the box. I have been doing some detailed performance analysis of the OS and hardware and have found that the box is adequate for the needs of the application.

However, my management and DBAs are not pleased with the performance of the machine. I would like to know if fellow HP-UX'ers out there feel like this is adequate hardware for this type of custom database? I know there are many factors that come into play here and trust me I've looked into many of them, but would like some formal proof that this machine is capable of supporting this environment.

The real problem lies in the application itself, with poorly written SQL and inefficient algorithms. I'm just trying to prove that the issue is *NOT* the hardware itself and need some type of proof other than the performance statistics I've been using, since that doesn't seem to be enough. Is there any documentation that states that the K370 with four CPUs installed is plenty of power to support 100 or fewer users in a simple socket transfer database environment? Any help would be appreciated. Thanks.

- Mike

[Steve Post]

I do NOT have any sure-fire way to "bless" the performance of the computer. I'm just an administrator like you. But here are some things you probably already looked at...
1. Look for disk bottle necks. If you have them, then it IS the computer making it slow, not the code.
2. Run one of the pieces of code, time it, fix it, then run it again. If the code runs 10 million times faster....there's your proof.
3. Look for memory problems with vmstat and glance plus.
4. Somehow look for high network load. Maybe the computer's fine. But people are overloading the network.
5. If indexes on the tables in sybase are set up inefficiently, you would have an impact.

I realize this is really vague. But I saw no one responded. I figured some unqualified hints are better than nothing at all.

Steve Post ^_^

[Philip Chan_1 · Accepted solution]

Hi Mike,

Your four CPU K370 should be sufficient to handle 100 users. The real problem that made such hardware looked bad is poorly written SQL statements. One symptom for poorly written SQL is that they will perform table scans rather than index scan hence large amount of disk I/O will be triggered. If you see from sar that the disk utilizations are high, say over 80% for most of the time, that mean the SQL programs were doing table scans.

These disk I/Os are the proof !!!

You MUST make your management be aware that poorly written SQLs are performance killer. Performance of a database system is always disk I/Os related.

Another area you should look into is to see if there were any live locks among different sessions. Occurrance of livelock is a result of poorly designed databases.

By upgrading hardware, you may gain 2-3 times performance, but turning a badly written SQL into its optimize form could speed up the system by 20 times EASILY.

Rgds,
Philip