what cache hit ratio are you getting for RD and WR

[skt_skt]

I have a V400 and F400 both reports write cache hit ratio below 40%. Also WR  misses are double than  the WR hits.

What do you observe with your systems? This is lesser than i expected  and the same time i dont have much insight in to how the write caching is handled in 3par.

please share your experiences/observations/thoughts

[skt_skt]

Or let me put it this way. Why do i see write service is higher than than read service time? Any one can share what goes behind on the 3par side which makes it this way.

[Dennis Handly]

What command are you using to get this data, statcmp?

"clihelp -col statcmp" will describe the columns.

A write cache hit means the block was already in a page in the cache.  If not, a new page must be allocated.

>Also WR  misses are double than  the WR hits.

You are doing random writes?

>I don't have much insight in to how the write caching is handled in 3PAR.

Data goes into the cache on at least two nodes and then is acknowledged.

>Why do I see write service is higher than than read service time?

If in the cache, it is just fetched.  For writes, it must be sent to two nodes.  Similar issues for writing to disk, depending on RAID level.

[skt_skt]

Glad to see some response coming up. The data is pulled using System Reporter.

what metrics would you look at 3par to determine what  percentage of write is random/sequential?

Mirroring writes is not really new the industry. Also raid penalty is applicable for all vendors.

>Data goes into the cache on at least two nodes and then is acknowledged.

due to the caching shouldn't the WR ACK should be received faster(smaller respnse time) than reading from the Physical Disk?

[Dennis Handly]

>The data is pulled using System Reporter.

(I only know CLI.)

>what metrics would you look at 3PAR to determine what  percentage of write is random/sequential?

I don't see any commands to do that.  You are suppose to know your host applications.  ;-)

>due to the caching shouldn't the WR ACK should be received faster (smaller response time) than reading from the Physical Disk?

I suppose if there are free cache pages.

[skt_skt]

what do you look under statcmp to check if cache pages are exhausting in the system/nodes?

[Dennis Handly]

Not too sure:

Queue Statistics  (A small number of Free and Clean)

Page Statistics (lots of CfcDirty)

[skt_skt]

DelAck( Delayed acknewledgment) for each drive types indicates the cache is unable to destage  to the disks as fast as the host writes coming in.