dm-devel.redhat.com archive mirror
 help / color / mirror / Atom feed
* dm-cache warming
@ 2013-07-26 15:24 Dave Pitts
  2013-07-29 10:16 ` Heinz Mauelshagen
  0 siblings, 1 reply; 3+ messages in thread
From: Dave Pitts @ 2013-07-26 15:24 UTC (permalink / raw)
  To: dm-devel

Hello:

I've been trying to do some performance testing for a client and can't seem to 
get the dm-cache warmed. The values displayed in the "dmsetup status" command 
seem too low. I've been trying to use the "raw" devices setup for the cache and 
NOT use a file system or database. I've tried sequential writes/reads to the 
devices (ie. dd ) and dividing up the space to be tested and do I/O over a 
number of segments to try to break any sequential I/O tests. Any ideas as to 
what I may doing wrong? Do I need to randomize the I/O?

System: RHEL 7.0 Alpha3, 3.9.7 kernel

Thanks in advance..

-- 
Dave Pitts             PULLMAN: Travel and sleep in safety and comfort.
dpitts@cozx.com

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: dm-cache warming
  2013-07-26 15:24 dm-cache warming Dave Pitts
@ 2013-07-29 10:16 ` Heinz Mauelshagen
  2013-07-29 10:40   ` Joe Thornber
  0 siblings, 1 reply; 3+ messages in thread
From: Heinz Mauelshagen @ 2013-07-29 10:16 UTC (permalink / raw)
  To: dm-devel


Hi,

dm-cache is a hot-spot cache. Ie. you need multiple accesses to the same 
block to get it promoted into the cache (depending on the policy plugin 
btw).

If your patterns is one-time-only access type, it ain't warm the cache.

Heinz

On 07/26/2013 05:24 PM, Dave Pitts wrote:
> Hello:
>
> I've been trying to do some performance testing for a client and can't 
> seem to get the dm-cache warmed. The values displayed in the "dmsetup 
> status" command seem too low. I've been trying to use the "raw" 
> devices setup for the cache and NOT use a file system or database. 
> I've tried sequential writes/reads to the devices (ie. dd ) and 
> dividing up the space to be tested and do I/O over a number of 
> segments to try to break any sequential I/O tests. Any ideas as to 
> what I may doing wrong? Do I need to randomize the I/O?
>
> System: RHEL 7.0 Alpha3, 3.9.7 kernel
>
> Thanks in advance..
>

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: dm-cache warming
  2013-07-29 10:16 ` Heinz Mauelshagen
@ 2013-07-29 10:40   ` Joe Thornber
  0 siblings, 0 replies; 3+ messages in thread
From: Joe Thornber @ 2013-07-29 10:40 UTC (permalink / raw)
  To: device-mapper development

On Mon, Jul 29, 2013 at 12:16:11PM +0200, Heinz Mauelshagen wrote:
> 
> Hi,
> 
> dm-cache is a hot-spot cache. Ie. you need multiple accesses to the
> same block to get it promoted into the cache (depending on the
> policy plugin btw).
> 
> If your patterns is one-time-only access type, it ain't warm the cache.

Remember there's the page cache in front of dm-cache too.  You may be
accessing data many times, but not actually hitting the disk with
your requests.

- Joe

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2013-07-29 10:40 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-07-26 15:24 dm-cache warming Dave Pitts
2013-07-29 10:16 ` Heinz Mauelshagen
2013-07-29 10:40   ` Joe Thornber

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).