topic Re: on increasing data segment size in Operating System - HP-UX

on increasing data segment size

Vishal Augustine — Mon, 15 Jul 2002 10:27:16 GMT

Hi,

What is the effect of increasing the data segment size on performance of a heavy loaded server ? What is the optimum value suggested ? Is it advicable to use EXEC_MAGIC and SHMEM_MAGIC ?

Thanks and Regards
Vishal

Re: on increasing data segment size

Trond Haugen — Mon, 15 Jul 2002 10:45:14 GMT

It depends on what kind of bottleneck you have. What ave you done to determine that?
The parameter only limits the mount of "data" each process can allocate so my guess is it won't help.

Regards,
Trond

Re: on increasing data segment size

Vishal Augustine — Tue, 16 Jul 2002 04:47:48 GMT

Hi,

malloc is failing at times. This means the server is heavily loaded. The ulimit of data segment is 65536 KB. I believe increasing the data segment size should solve the problem. Using EXEC_MAGIC flag while linking will also solve the issue. Am I right ? What I don't know is its impact on the performance of the server (both on the machine and individual server-processes)

Thanks and Regards
Vishal

Re: on increasing data segment size

Bill Hassell — Tue, 16 Jul 2002 11:11:40 GMT

malloc failing does not indicate a heavily loaded server. malloc will fail for three basic reasons: the request exceeds the kernel fence: maxdsiz, or the reuest exceeds 940 megs, the limit for standard executables (where multiple quadrant maps are not available, ie, the MAGIC executables), or you are out of swap space. Unfortunately, the program cannot determine the real reason, so you have to investigate each scenario.

Start with maxdsiz. The default is 67 megs which is tpyically way too restrictive. Change this fence value to 940 megs for standard programs, or bump it up to 1750 megs if you plan on creating MAGIC executables.

Then run swapinfo -tam when malloc fails to see if most of the swap space is used (not just reserved). If all is well, then use chatr to change you executable(s) to map quadrants 1 and 2 together, thus allowing up to apx 1750 megs in the data area.

The details are in: /usr/share/doc in the two whitepapers: mem_mgt and proc_mgt.

Re: on increasing data segment size

Trond Haugen — Wed, 17 Jul 2002 05:42:10 GMT

Jut to add to Bill's answer. You should not experience any change in performance when doing these changes. But as always with tuning only doing it will tell for sure.

Regards,
Trond