Measure User's Guide

Balancing and Tuning a System
Measure User’s Guide—520560-003
7-16
Checking and Tuning Problem Areas
°
Small files. By providing enough cache to hold the entire file, you can avoid
many disk I/Os and so improve performance. Providing cache does not
guarantee that the file remains in cache. Unless the small file is busy, its blocks
are likely to be swapped out in favor of busier files.
Use the STATISTICS and DETAIL clauses of the FUP INFO command to examine
the block sizes and index levels of the major files on each disk.
This text provides guidelines for sizing cache for DP2 disks. Adjusting disk cache can
have a major effect on system performance. Do not change cache sizes unless you
have a problem.
•
Although you configure cache only for the primary disk process, the system
automatically configures the same amount of cache for the backup disk process.
Both CPUs are affected.
•
When examining the file activity on a disk, consider only heavily used files. Briefly
used files do not influence performance and should not influence cache sizing.
Disk cache consists of four separate caches, each containing blocks of different sizes
(512 bytes, 1024 bytes, 2048 bytes, and 4096 bytes). RVUs prior to D40.00 provide a
number of blocks for each cache that equals the number of PINs multiplied by 7 and
rounded up to a full-page boundary. D40.00 and later RVUs provide a maximum of 56
blocks for each cache. Thus, when configuring disk cache, consider that the impact on
CPU physical memory includes all four caches for each disk process, primary and
backup, in the CPU.
When configuring the cache sizes for each disk, consider the file activity (as discussed
at the beginning of this section) in each of the four caches for that disk. The cache a
file uses is based on the file block size. Examine the block size of the major files on the
disk by using the ENDING-FREE-BLOCKS counter of the DISK entity. (Alternatively,
you can use the STATISTICS or DETAIL clause of the FUP INFO command.)
Examine the cache statistics collected in the Cn counters of the DISC entity. These
counters can help you size the cache:
•
FAULTS. A cache fault occurs when the disk process expects to find a block in
cache but discovers that the memory manager has removed it. Cache faults are a
result of contention between the CPU memory manager and the CPU disk
processes. A high FAULTS value (greater than 5) might be caused by several
things:
°
The disk cache on the CPU might be too large.
°
Too many processes in the CPU might be locking down too many memory
pages for the PFS for each process.
°
Too many system processes might be locking down memory for resident code
and data pages.
°
Too little physical memory might be available to the CPU.