Sybase Perfomance Problems

[Jeff Machols]

We are seeing a very high number of system calls from our Sybase 11.9.2 servers on HPUX 11.0. We do not appear to have the same issue on 11.0.3. During peak times we are seeing over 200,00 select calls per second per dataserver process! Has anyone seen anything like this?

[James Murtagh]

Hi Jeff,

That is a lot of syscalls, however this will depend on the amount of CPUs you have, please let me know. What you are refering to is a known problem with select called a "thundering herd" syndrome - the syscall will "poll" file descriptors waiting for an event to happen. When this occurs all selects waiting will be awakened - hence causing very poor performance. The code has been drastically changed on 11i and I haven't seen any issues with it on that or any of its subsequent releases (they all use the same code base after all). See the following link on the parameter to check and further analysis:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x4b8f3f9a15add711abdc0090277a778c,00.html

Cheers,

James.

[Jeff Machols]

We currently have 6 CPU's and about 30 sybase servers running, which I know is not the best setup.