Nike disk array Cache size

[Team Leader UNIX DBA]

I am trying to determine if it is justify increasing the cache size of the Nike disk array from 10M to 64M.

My HPUX OS buffer cache is 600M in size and is averaging 100% for %rcache or GPL_MEM_CACHE_HIT_PCT, also 97% %wcache. It looks like Nike cache is redundancy if the OS buffer cache is large enough like the box that I have. Given in theory most of the data in the Nick cache should be in the buffer cache anyway and most of the buffer cache miss would generate a Nick cache miss therefore a physical disk access? That sound having a large Nick cache pointless. Am not saying it is totally unnecessary assuming the storage processor has its own algorithm to do certain read ahead and keep the data in the Nike cache. Beside of that, would there be a significant performance gain to upgrade the cache in the Nike from 10M to 64M?

Thanks for your help,
Keith Leung

[Michael Tully]

Hi,

I haven't got the piece of hardware, but there
has been many posts recently in regards to the
usage of the kernel buffer cache. General
consensus is that having any more than 300Mb is
a pointless exercise. This could be what is
causing your current problem. Do a search on
buffer cache in the forums here and you will see what I am talking about. Below are some:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xf49203bbece8d5118ff40090279cd0f9,00.html
http://forums.itrc.hp.com/cm/QuestionAnswer/1,11866,0x997cabe92dabd5118ff10090279cd0f9,00.html

[Bill McNAMARA_1]

depends on your raid level.

If you're in R5, I'd suggest assigning more memory for the calculation space around 2M per LUN.

You can turn on statistics on the nike via gridmgr, but I'd prefer to rely on host based tests and turn ALL stats off on the Nike. (Bet you have them on)

Most perf gain on the Nike will come from elsewhere. (bet you've messed up that config)
See my posts here:
http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xd5178ffa98a2d5118ff10090279cd0f9,00.html

Later,
Bill