Write Cache Hit 100% - MD3220

Question

We run a PowerVault MD3220, 24 600GB 10K SAS Toshiba derives, uses SAS cables to attach to 2 256MB RAM clustered servers running W2k12 Hyper-v. This is a fairly new setup for us, and runs about 40-odd VMs, ranging from Terminal Servers and Virtual PCs to light SQL and Exchange.

A client has complained about performance in her Windows 7 VM. I've noticed a certain sluggishness in some machines as well, though there has been no widespread complaints. I have used the built-in monitoring and found the write cache hit % is always 100%. The read Cache is 50-70%.

I can't really find anywhere that explains to me what Write Cache % means. Is 100% bad and requiring corrective action, or is it normal and means everything successfully caches?

From the server POV, disk queue seem to be <1 consistently, but I am not sure that is an accurate reflection of what is going on on the DAS box.

My Cache Block size is 4KB - if the above does indicate issues, would that be a sensible parameter to try tuning?

I'm aware that's a few questions - if there is a dummy guide to performance measuring/tuning on the MD3220, it would be great to be pointed at it.

Many thanks for any help.

Nick

DELL-Sam L · Answer

Hello officeanyplace,

Here is the deployment guide for the MD3220. ftp://ftp.dell.com/Manuals/all-products/esuprt_ser_stor_net/esuprt_powervault/powervault-md3200_Deployment%20Guide_en-us.pdf You didn’t list how many connections that you have from each server to the MD. If you only have a single connection from each controller on the MD 3220 to the server then that can cause your write cache at 100%. Also you stated that you both servers are running with 256MB of ram & running Hyper-v. this can also be an issue as 2012 requires a minimum of 512MB of ram. Also running about 40 VM on 2 servers can also be creating a bottle neck as well. Also what raid level are you running for your virtual disks?

Please let us know if you have any other questions.

officeanyplace · Answer

We have 2 connections from each server to the Md3220. I must have mistyped, each server is running with 256 GB of RAM, not MB. I am running Raid 10.

It may well be that the VMs are too much for the disk array (There doesn't appear to be any issues on the host servers) - what I'm trying to find out is how I can understand the metrics available to confirm that, and to work out if buying/enabling the High Performance option would be helpful to me, for example.

DELL-Sam L · Answer

Hello officeanyplace,

The main tools that you can use to test the MD with is IOMeter or SQLIO. Both are free & can be downloaded from the internet.

Please let us know if you have any other questions.

BenTx · Answer

For anyone else looking at this metric, you want the write cache hit percentage to be 100%. It shows what percent of write operations are completed from cache. If this number is below 100% it indicates cache performance issues.

The MDSM help windows are actually pretty good. The Help button in that window will give an explanation of each metric.

Write Cache Hit Percentage: The percentage of all write operations that are handled in the cache. This value includes both sequential and nons equential write operations.

officeanyplace · Answer

Thanks very much for this Ben, that's what I needed. My write cache hit % goes down to something like 99.99% - is that an issue, in the context of 'beloe 100% it indicates cache performance issues', or is that so negligible as to not worry?

I have a vague memory of looking at that help text and simply not understanding it - your rewording helped.

PowerVault

Write Cache Hit 100% - MD3220

Was this post helpful?