STARWIND WB mode has caused me everything

sunyucong · Fri Aug 10, 2012 10:52 pm

Doing a simple starwind restart today, and after reboot, the volume that configured with WB has lost all the data in the cache, result horriblly corruputed. the volume with write-through is totally fine.

Have been doing this for a while and finally, it bite me hard. I guess that's it, all my data is gone now.

Fri Aug 10, 2012 11:21 pm

Write-back mode is not expected to be used with a single head (non-HA). And HA does not mean you should not use backups.

DUOK · Sun Sep 02, 2012 5:24 pm

So it is unsafe to use the write-back cache in single head (non-HA)?
In Starwind v6, is also unsafe to use the write-back cache in single head?
Assuming, of course, that the raid controller have a battery, and the server is connected to a UPS.

Another question, in v6, is there not the possibility to change the expiration time of the cache? why?

Thanks,

Sun Sep 02, 2012 9:25 pm

We don't control operating system so it's not safe to use Write-Back cache with anything on a single node. We'll put a disclaimer with a next update. Also we'll have cache expiration control mode. It's legacy setting so far.

DUOK wrote:So it is unsafe to use the write-back cache in single head (non-HA)?
In Starwind v6, is also unsafe to use the write-back cache in single head?
Assuming, of course, that the raid controller have a battery, and the server is connected to a UPS.

Another question, in v6, is there not the possibility to change the expiration time of the cache? why?

Thanks,

gstephenson · Mon Sep 03, 2012 2:41 pm

What about the scenario where if we have 2-node HA and use the 'mark as synchronized' feature when one node is down -- does the remaining target switch to write-through automatically?

Graham

Mon Sep 03, 2012 8:33 pm

Yes, 2-way synchronous mirroring configs switch to write-thru from write-back and flush cache immediately after they lose partner. That's why we recommend using 3-way HA as it does not suffer from performance degradation and is much more reliable (military and aerospace grade):

http://www.starwindsoftware.com/why-3-w ... than-2-way

(sorry it's locked by subscription but I guess you have an account and we'll open this PDF in any case)

gstephenson wrote:What about the scenario where if we have 2-node HA and use the 'mark as synchronized' feature when one node is down -- does the remaining target switch to write-through automatically?

Graham