StarWind Software

Hi all,

Some advice will be appreciated. I have two different server types that I want to use in a HA setup. One is a Intel 12bay, and another a SuperMicro 24 bay(only 12 populated).

The Intel has an Adaptec 5405Z 3GB SAS/SATA controller with 12 3TB 6G SATA disks attached in RAID10(I know the disks will downgrade to 3G due to the controller) This will be the secondary unit in HA.

Then in the SMC I will have a LSI 9280 6G Controller with 12 2TB 3G SATA disks. Again this array will only run on 3G as the disks are the limitation here. However later I will add 6G disks.

As you can see I have a bit of a mix here and obviously 6G controller with 6G disks will be better, however the Supermicro is in use at the moment.

So, the question is this. If the Primary SAN is "faster" than the Secondary SAN, will it drag down the performance of the HA target??? I assume with HA it waits for confirmation of the write on the secondary server before marking the write as completed (Syncronous replication) ?

Or should I try to keep the RAID controllers the same if possible, and have the same performce on both SANs?

To add to this, one will be a Intel processor and the other AMD, but I assume this wont be an issue. Both will have plenty of RAM.

Ideas will be appreciated.

Squadron speed is limited with the speed of a slowest ship in a squadron. So in your case with caching turned OFF or cache used with write-thru policy you'll have resulting performance being ~1/2 (or sometimes closer to 40-45%) of the performance slowest HA node has. At the same time if you have write back cache enabled disk subsystem performance should be pretty much approximated. Unless you have real heavy write load (non-linear video editing or sound/visual content capture) rendering write cache useless.

So you're saying that if I use write-back cache, I should still get decent performance as the targets will be used to serve VM images, so heavy random data.

Pretty close to "YES".

hfourie wrote:So you're saying that if I use write-back cache, I should still get decent performance as the targets will be used to serve VM images, so heavy random data.

thanks for the reply.

You're welcomed! Please let us know and provide some feedback when you'll give listed scenario a try. Thank you!

hfourie wrote:thanks for the reply.