StarWind Software

So I've been having issues with Starwind performance being, well pretty bad and I have stumbled across an interesting discovery.
fyi Product is Native San for Hyper-V, 2 nodes, both severs are dell pe720 with 192GB ram, 6 disk 840 Samsung Pro in Raid 10, 10gbit connection for sync, separate 10gb for iscsi.


What I found was that if I created a ramdisk on server A, and connected to it via the MS iSCSI iniator, and ran ATTO benchmark (or any other benchmark for that matter) performance was lower than a local mechanical hard disk! BUT!!!! If I connected the same ram disk to Server B over the 10gbit link, the performance was MUCH higher. Please see the attached photo.


So my dilemma is....how do I fix this? I do not want to have to spin up 2 additional servers just for storage, that was the whole point of using the "Native" SAN for Hyper-V.

Just as an additional note- same performance issues with the SSD array-

And here is the SSD array without using Starwind. Quite a bit of difference.

TCP loopback within Windows is SLOW. We've applied special optimization technique to boost loopback performance dramatically. This will be included with the upcoming version. If you'd like to test it now (it's a stand-alone component so no need to change anything) we can "leak" you premature version.

Well I had this on Server 2012 beforehand with similar issues, and I was under the impression that Server 2012 had much much faster loopback as it never made it all the way to the TCP/IP stack like it does in server 2008.

But if you want to 'leak' me something to see if it will fix it I will be more than happy to try it out! This gear isn't in production yet and I am going to keep it that way until we have everything ironed out

Yup. 2012 is faster and different. But you need to "tweak" it. Unfortunately it's not just a registry setting or whatever. See this:

http://blogs.technet.com/b/wincat/archi ... -path.aspx

Drop a line to support@starwindsoftware.com with a link to this thread and guys will help with a test package.

eickst wrote:Well I had this on Server 2012 beforehand with similar issues, and I was under the impression that Server 2012 had much much faster loopback as it never made it all the way to the TCP/IP stack like it does in server 2008.

But if you want to 'leak' me something to see if it will fix it I will be more than happy to try it out! This gear isn't in production yet and I am going to keep it that way until we have everything ironed out

anton (staff) wrote:Yup. 2012 is faster and different. But you need to "tweak" it. Unfortunately it's not just a registry setting or whatever. See this:

http://blogs.technet.com/b/wincat/archi ... -path.aspx

Drop a line to support@starwindsoftware.com with a link to this thread and guys will help with a test package.

eickst wrote:Well I had this on Server 2012 beforehand with similar issues, and I was under the impression that Server 2012 had much much faster loopback as it never made it all the way to the TCP/IP stack like it does in server 2008.

But if you want to 'leak' me something to see if it will fix it I will be more than happy to try it out! This gear isn't in production yet and I am going to keep it that way until we have everything ironed out

Well that makes sense. Well 2012 seems to have a littany of issues any way so I am not surprised it takes more to get it to work. You can't even back the thing up with their own product (DPM) if you use CSV....

Email sent. I hope this resolves my issue. I couldn't believe how fast it was to a remote server when I took the local out of the mix. If I can have TWO connections that fast then I will be in business!

Thanks!

Just got an update from our R&D, looks like the loopback accelerator will be available for download as soon as next week.
I'll keep you posted!

As promised - here is the link to the new build with loopback acceleration
http://www.starwindsoftware.com/tmplink ... 130326.exe
No special update steps, install on the first server, synchronize and do the same on the second one.