HP NonStop G-Series - Manual (Page 153)

Balancing and Tuning a System
Measure User’s Guide — 520560-003
7- 17
Checking and Tuning Problem Areas
°
Backup disk processes might be consuming amounts of cache larger than their
primary memory (TMF and control points).
Depending on the cause of the cache faults, you might choose to reduce the
amount of cache configured for the disk, reduce the amount of cache configured in
the CPU, move processes out of the CPU, or reconfigure system processes.
•
BLKS (blocks allocated). If the disk has far more blocks allocated than it has in
use, the amount of cache configured for this disk is too large. Determine which of
the caches to reduce by considering the file activity in each cache.
•
Counters reflecting hit/miss ratios. When you examine the hit/miss ratio for a
particular cache, note the file activity for that cache. For example, a cache with a
substantial amount of sequential I/O activity should have a good hit/miss ratio. A
cache with a substantial amount of random I/O activity has a chance of producing
a good hit/miss ratio. A cache with a substantial amount of unbuffered write activity
cannot produce any cache write hits.
In addition, note the file activity of the other caches. For example, if all file activity
for the disk is in one cache, give that cache block size the most memory. If the file
activity for the disk is split between two caches, consider which files can benefit
most from a larger cache.
The DISC counters that reflect cache hits and cache misses are:
In general, for each CPU, configure DP2 disk cache so the READ HITS and WRITE
DIRTYS for each of the four caches for each disk process (primary and backup) in the
CPU are as high as possible while the CACHE FAULTS for each disk process and the
Measure SWAPS counter for the CPU remain low.
To get more information for cache sizing on systems running D-series RVUs, use the
PUP LISTCACHE command with the STAT option, initializing the counters with the
INIT option. Initialize the counters any time you begin a measurement—after modifying
the cache configuration, before a peak period, and so on.
DISC Counter The percentage of...
CACHE READS Cache I/O operations that were reads
CACHE READ HITS Cache read operations that found the requested block in
cache
CACHE READ MISSES Cache read operations that did not find the requested block
in cache
CACHE WRITES Cache I/O operations that were writes
CACHE WRITE DIRTYS Cache write operations that found the requested block in
cache and dirty (saving a disk I/O)
CACHE WRITE CLEANS Cache write operations that found the requested block in
cache and clean
CACHE WRITE MISSES Cache write operations that did not find the requested
block in cache