Re: Performance advice

sumGirl wrote: > NETTYPE soctcp,2,150,CPU > NETTYPE ipcshm,2,150,NET CPU processes don't do soctcp very well, and NET processes don't do ipcshm very well. swap this to NETTYPE soctcp,2,150,NET NETTYPE ipcshm,2,150,CPU > RESIDENT 0 # Forced residency flag (Yes = 1, No = If you've got so much memory - ENSURE it's pinned down: RESIDENT -1 > NOAGE 1 # Process aging > AFF_SPROC 0 # Affinity start processor > AFF_NPROCS 0 # Affinity number of processors No age is Good for many processes, but affinity is Good too (pending it's support in AIX which I would expect) AFF_SPROC 1 (1 or two - use the last 3 cpu's) AFF_NPROCS 3 The reason i suggest using the last 3 cpus (using whatever numbering scheme AIX uses) is that on some platforms, processor zero does specific kernel and/or hardware tasks, so don't interfere with responsiveness on that. > BUFFERS 150000 # Maximum number of shared buffers like the other's said.... > NUMAIOVPS 2 # Number of IO vps I hope you are using KAIO. If not..... > CKPTINTVL 1800 # Check point interval (in sec) Pointless. Return to 300, especially if you are adding yet more buffers as recommended. > # Read Ahead Variables > RA_PAGES 128 # Number of pages to attempt to read > RA_THRESHOLD 120 # Number of pages left before next Excessive. That will impact on other threads. Drop to 32,28. Prove that it needs to be higher before going mad on this one. Is your system OLTP or warehousy? > DBSPACETEMP work1dbs,work2dbs # Default temp dbspaces is one and only one of these a T (temp) dbspace?

"Andrew Hamm" <ahamm@mail.com> wrote in message news:c61qsi$6v3ch$1@ID-79573.news.uni-berlin.de... > > No age is Good for many processes, but affinity is Good too (pending it's > support in AIX which I would expect) > Well, I'd be interested in hearing someone whose views I respect expounding the benefits. But perhaps you could lead off, Andrew .... ;-) > > DBSPACETEMP work1dbs,work2dbs # Default temp dbspaces > > is one and only one of these a T (temp) dbspace? Why?

Neil Truby wrote: > "Andrew Hamm" <ahamm@mail.com> wrote in message > news:c61qsi$6v3ch$1@ID-79573.news.uni-berlin.de... >> >> No age is Good for many processes, but affinity is Good too (pending >> it's support in AIX which I would expect) >> > > Well, I'd be interested in hearing someone whose views I respect > expounding the benefits. But perhaps you could lead off, Andrew .... > ;-) once again, may depend on the architecture, and I don't know AIX, but the two main benefits of noage and affinity are: 1) pinning a process to a CPU saves the cost of migration to another CPU, which might involve relatively expensive memory copying. 2) pinning AWAY from the first physical CPU may be good for kernel and/or hardware processing 3) noage is good 'cos it give IDS a lot of guaranteed CPU time that is typically needed. If there is nothing else happening on the machine apart from clients (4GL or java or whatever) then they are more than amply serviced by the remaining CPU. Most of the work on a dedicated database machine is in the engine. 4) there is no point 4 > >>> DBSPACETEMP work1dbs,work2dbs # Default temp dbspaces >> >> is one and only one of these a T (temp) dbspace? > > Why? Because there's only two spaces listed :-) Of course, more spaces would be nice, and a decent mix of types in amongst that, but some things are too fussy to go into detail in the first reponse - I'm trying to cut down...

Andrew Hamm wrote: > sumGirl wrote: > > NETTYPE soctcp,2,150,NET > NETTYPE ipcshm,2,150,CPU I missed one. You have 3 CPUVP's allocated, so change the ipcshm to: NETTYPE ipcshm,3,100,CPU because then the load applied to the CPU will be better balanced. Art S. Kagel has expounded on this subject quite admirably in the past. I'd actually be interested to know if this point has been corrected in the engine code these days. Any thoughts Art? Any recent experiments applied to see if it still matters?