StarWind Software

Hi, I was testing Native SAN for Hyper-V with RAID5 array of SSD's and huge performance loss with test on iSCSI device vs physical device. After testing all real and not real possibilities I decieded to make some test with iSCSI on local PC - same result.

Then I started testing Starwind on another versions of Windows:
Windows 2012 with raid10 array: Same performance loss.

Windows 2008R2 with SAS HDD's w/o array: No any losses on performance.

Should I try to setup windows 2008 on my main configuration with RAID5 SSD's? Or maybe problem in RAID? Or or you know some recipe to make windows 2012 to work nicely with Starwind (don't want to degrade already purchased licences)?

1) Replace your benchmark with Intel I/O Meter (long) and ATTO Disk Benchmark (short).

2) Don't use parity RAIDs in productions. They are SLOW (unless used with special file systems N/A for Windows yet). Also they KILL flash. So stick with RAID10.

3) Use Windows Server 2012 and don't forget to enable loopback acceleration with vSAN (Native SAN) for Hyper-V.

4) Report updated results telling more about your connfig (network interconnect topology). We'll continue.

anton (staff) wrote: 3) Use Windows Server 2012 and don't forget to enable loopback acceleration with vSAN (Native SAN) for Hyper-V.

quick question: what steps are there to enable loopback acceleration? Or is this always enabled when using the latest Starwind version?

It's part of vSAN (Native SAN for Hyper-V). There's no loopback so no acceleration for bare metal StarWind and non-Windows Server 2012 configs.

Ara wrote:

anton (staff) wrote: 3) Use Windows Server 2012 and don't forget to enable loopback acceleration with vSAN (Native SAN) for Hyper-V.

quick question: what steps are there to enable loopback acceleration? Or is this always enabled when using the latest Starwind version?

Okay, let's make it clear about testing configuration.

I have two identical servers on Supermicro 1027r-WRF platform with 2xIntel Xeon E5-2620 CPU, 80GB RAM, 4x256GB SSD's, networking based on two integrated Intel I350 controllers (one for external, one for Heartbeat sync) and one Intel x540 10GB NIC with two VLANs (one for internal synchronization channel, one for client connections).

Now I recreated RAID on one of them and make RAID10 with fresh and clean Windows Server 2012 installation (just Windows with Hyper-V role, drivers for NIC's and Storage). Test configuration:

Just physical drive - on the left
Starwind iSCSI virtual disk w/o cache, w/o multipath connected to host computer with MS iSCSI initiator (no networking involved, all on 1 local PC) - on the right Yes, perfomance is not so bad, but it more that two times slowly than physical drive.

Next idea: what if problem in MS iSCSI initiator? -Tested with Starwind initiator - results same.

If you want, I can even give you remote access to my server, if you wanna test it by yourself.

Second server now have 2008R2, clean and fresh, just Windows and drivers, tests:

1. Just physical drive - on the left. Starwind iSCSI virtual disk w/o cache, w/o multipath connected to host computer with MS iSCSI initiator - on the right. 2. Just physical drive - on the left. Starwind iSCSI virtual disk w/o cache, w/o multipath connected to host computer with Starwind iSCSI initiator - on the right. (enabled raid controller cache)

May I ask what StarWind device type have you used for a test (i.e. Deduplication, Basic Image, HA, etc.)?

Basic Image

anton (staff) wrote: 3) Use Windows Server 2012 and don't forget to enable loopback acceleration with vSAN (Native SAN) for Hyper-V.

Where do you enable this? I can't find any settings in the Console...

kevrags wrote:

anton (staff) wrote: 3) Use Windows Server 2012 and don't forget to enable loopback acceleration with vSAN (Native SAN) for Hyper-V.

Where do you enable this? I can't find any settings in the Console...

I think it's in installer:

Yes, it is installed. Can you confirm that you are using 127.0.0.1 IP address to connect to Basic Image?