MAIN

ROOTSIZE

Logical Logs + [Phys logs] + [Temp Tables] + [Data] + [On-Arc.catalogs] + crtl info (res. pages). This is a basic size calculation formula. For systems of any import most of these logical units should of course be on separate spindles.

SERVERNUM

unique for each engine on CPU.

CPU USAGE

RESIDENT

On a dedicated server turn this on (1). The parameter is ignored if it is not supported.
In a mixed environment turning this on may make Informix "not play well with it's freinds"

NUMCPUVPS *

#HWCPUs - 1 if #HWCPUs > 3
#scan thrds (frag issue) SB a multiple or factor of NUMCPUVPS determine w/ glo ath rea.
For data loading = #HWCPUs unless HPL used, then = HWCPUs – CONVERTVPS (onpload config param)
I have seen systems that enjoy 1:1 VCPUS to HW CPUs relationship and I have seen some that prefer 2:1:1

MULTIPROCESSOR

SINGLECPU

Both of these turn on different housekeeping mechanics. If NUMCPUVPS is set to one it is very important to turn both of these off, or 0 and 1 respectively (yech). You will see no benefit to configuring one CPUVP unless these are both off. This is overlooked more often than not!
I would suggest that these should really be considered as one parameter together as if one is adjusted the other should be adjusted likewise.

NOAGE

1 if supported this parameter turns off Unix nicing (this is good). Nicing is the Unix mechanism that lowers a processes priority over time to ensure equality in a mixed environment.  As Informix just wants machine cycles a dedicated server should have this enabled.
Tuning for a mixed environment is not so black and white. It may be best to have this off initially and turn it on if it is desired to give Informix more CPU time.

AFF_SPROC

CPU number to start binding to. mpstat will provide number of CPUs and their associated number. This numbering system seems whimsical.

AFF_NPROCS

Number of hardware CPUs to bind to.
NUMCPUVPS = HW CPUs – AFF_SPROC.
This can be very beneficial for systems that are not dedicated to the DB.
Other non-Informix processes may be allowed to run on the HW CPUs identified here, but the CPUVPS will be restricted to those identified.

USEOSTIME

0 internal timer is faster. I have never seen OS timing used.
NOTES:

-g glo There should be a 10:1 ratio between time spent in usr vs. sys for CPUVPS. (This has become perhaps 3:1 with KAIO). If sys is too high on a system with just a few HW CPUs try using just 1 CPUVPS.

-g rea If this shows seven or more threads waiting adding CPUVPS can bring it down.

UNIX COMMANDS:

mpstat

sar

uptime

Very marginal tool. load average is a combination of system resource measurements. I have seen slow systems with a la of 2 and I have seen systems that seemed find at double digits.
Useful only relative to an earlier measurement on the same machine.

 

 

 

Introduction

Main & CPU

Disk IO

Logging

Memory & DSS

Miscellaneous

Notes

Print Version

bamph

                     

Web design and content bamph