On Nov 7, 2007 6:09 PM, Brian Nielsen <[EMAIL PROTECTED]> wrote:

> I'm working on fine tuning my balance between MDC hits, VM paging rate,
> Linux swap rate, and Linux guest use of memory for cache.  The one piece
> of data I'm missing is the hit ratio of the Linux cache.  My research so
> far has not turned up anything that reports this value.  Anybody know if
> this statistic is even available (SLES 9 & 10) and, if so, where to find
> it?

There's no such metric right now. Future kernel levels have added a
tracing facility in the block device layer, and proper processing of
that can compute cache hit ratio (among others) from that trace. I'm
in the process of building that gear.

Actually, believe it or not - I just booted the system with the things
in place for the first time. Analysis of a small portion of trace
shows things like this:
CPU0 (94,1):
 Reads Queued:         424,   11,332KiB  Writes Queued:         146,      620KiB
 Read Dispatches:      402,   11,372KiB  Write Dispatches:       87,      620KiB
 Reads Requeued:         0               Writes Requeued:         0
 Reads Completed:      402,   11,372KiB  Writes Completed:       87,      620KiB
 Read Merges:           24,      284KiB  Write Merges:           59,      236KiB
 Read depth:             6               Write depth:             6
 IO unplugs:           382               Timer unplugs:          20

Need some more digging to understand the numbers and see what I can
get out of that.

Rob
-- 
Rob van der Heij
Velocity Software, Inc
http://velocitysoftware.com/

Reply via email to