Thread
-
Re: [HACKERS] Neverending query on 6.5.2 over Solaris 2.5.1
Tom Lane <tgl@sss.pgh.pa.us> — 1999-10-22T15:33:56Z
Fernando Schapachnik <fpscha@ns1.via-net-works.net.ar> writes: >> enough buffer space to avoid thrashing. You need to increase your >> kernel's max-shared-memory-block-size (SHMMAX, I think) parameter >> so that you can run with a more reasonable -B setting. A lot of >> kernels ship with SHMMAX settings that are ridiculously small for >> any modern machine. > What value would you advise for shmmax? Well, with the default number of buffers (64) Postgres requires about a megabyte (I think a tad over 1Mb, in 6.5.*). Extra buffers are 8K plus a little overhead apiece. If you are running with more than a couple of active backends at a time then you probably want to use more than the default number of buffers. But I have no advice on how many is appropriate for what size of installation --- can anyone else help? regards, tom lane
-
Re: [HACKERS] Neverending query on 6.5.2 over Solaris 2.5.1
Brian E Gallew <geek+@cmu.edu> — 1999-10-22T15:57:11Z
Then <tgl@sss.pgh.pa.us> spoke up and said: > Fernando Schapachnik <fpscha@ns1.via-net-works.net.ar> writes: > > What value would you advise for shmmax? > > more than the default number of buffers. But I have no advice on > how many is appropriate for what size of installation --- can anyone > else help? Unless you are severely resource constrained, think big. Generally speaking, the kilobytes of memory you'll lose to kernel structures are irrelevant. Performance is generally not an issue, either. -- ===================================================================== | JAVA must have been developed in the wilds of West Virginia. | | After all, why else would it support only single inheritance?? | ===================================================================== | Finger geek@cmu.edu for my public key. | =====================================================================