v6 NAS question

awedio · Wed Aug 29, 2012 12:13 am

Since v6 now supports Clustered Scale-Out NAS:

Using Server 2012, does the NAS support:
1) SMB 3.0
2) Hyper-V VMs from the file share

Wed Aug 29, 2012 5:00 am

Yes, absolutely. That's was one of the scenarious it was created for.

awedio · Mon Sep 03, 2012 1:15 am

If v6 is used as NAS, is this kinda what the architecture would look like? With iSCSI replacing the FC?

Image is provided courtesy of http://www.windowsitpro.com/article/win ... san-143844

Mon Sep 03, 2012 4:29 am

Yes and no. Yes, you can do it but you'll have to configure SoFS manually. Here we are:

http://www.starwindsoftware.com/sw-prov ... le-servers

Also you can skip using FC or iSCSI as a back end and use SoFS directly running StarWind on top of them. So you don't have physical shared storage. Here we are:

http://www.starwindsoftware.com/sw-conf ... le-servers

And here are some feedback from people using this design efficiently:

http://social.technet.microsoft.com/For ... cee3c0c70d

Built-in NAS configurator allows you to create simple NAS (CIFS/SMB or NFS) failover cluster on top of StarWind so you'll be able to either
provide HA file share to your users or to keep VHD images and run Hyper-V from it (as what you wanted). But SoFS is preferred here as it allows to have som extra benefits (SMB Direct, pseudo-multipath based on DNS and so on).

awedio · Mon Sep 03, 2012 7:07 am

Anton, thx for the links. I'll start reading & studying asap!

Mon Sep 03, 2012 7:10 am

My preferred scenario is - StarWind running directly on SoFS servers. If you cannot manage skip using them and cannot install StarWind directly on Hyper-V nodes (w/o any external hardware).

awedio wrote:Anton, thx for the links. I'll start reading & studying asap!

awedio · Mon Sep 03, 2012 7:23 am

On the Technet forum VR38DETT (Starwind employee) said

V6 is out and it can do both. You may however will to wait for upcoming one with log-structuring to fight I/O blender and flash level cache. Fow now dedupe is slowing down I/O but it will increase IOPS.

Anton,

Can you share more info on "..log-structuring to fight I/O blender and flash level cache"?

Mon Sep 03, 2012 7:29 am

Sure. Upcoming version of StarWind (the one we have under heavy development) has log-structured file system. Component completely eliminating random writes killing virtualized environments.
It's VM I/O-aware layer, something similar to what Virsto and Tintri do. It does all writes in sequential way with a huge pages (good for big stripe RAIDs and also good for SSDs). And we'll add flash memory as L2 cache to already existing L1 cache (RAM).

awedio wrote:On the Technet forum VR38DETT (Starwind employee) said
V6 is out and it can do both. You may however will to wait for upcoming one with log-structuring to fight I/O blender and flash level cache. Fow now dedupe is slowing down I/O but it will increase IOPS.
Anton,

Can you share more info on "..log-structuring to fight I/O blender and flash level cache"?

awedio · Mon Sep 03, 2012 7:31 am

anton (staff) wrote:My preferred scenario is - StarWind running directly on SoFS servers. If you cannot manage skip using them and cannot install StarWind directly on Hyper-V nodes (w/o any external hardware).

You mean using the NAS configurator & have Starwind as "backend" storage for the FS?

awedio · Mon Sep 03, 2012 7:34 am

anton (staff) wrote:Sure. Upcoming version of StarWind (the one we have under heavy development) has log-structured file system. Component completely eliminating random writes killing virtualized environments.
It's VM I/O-aware layer, something similar to what Virsto and Tintri do. It does all writes in sequential way with a huge pages (good for big stripe RAIDs and also good for SSDs). And we'll add flash memory as L2 cache to already existing L1 cache (RAM).

awedio wrote:On the Technet forum VR38DETT (Starwind employee) said
V6 is out and it can do both. You may however will to wait for upcoming one with log-structuring to fight I/O blender and flash level cache. Fow now dedupe is slowing down I/O but it will increase IOPS.
Anton,

Can you share more info on "..log-structuring to fight I/O blender and flash level cache"?

The logs can use dedicated SSDs (like Virsto). Will this be v7?

awedio · Mon Sep 03, 2012 7:37 am

Another post from VR38DETT

Take a look at DataCore as another good example of properly implemented caching thing (they even do use polling stealing CPU cycles in favor of faster I/O response) under Windows. Solaris/FreeBSD have ZFS with multi-level (RAM and SSD, something we're still missing...) ARC. That thing shines

Why is Starwind missing "multi-level ARC" (like ZFS)?

Or is this part of the upcoming version (the one you say is under heavy development)

Mon Sep 03, 2012 7:53 am

We'll release flash as a second level cache pretty soon.

awedio wrote:Another post from VR38DETT
Take a look at DataCore as another good example of properly implemented caching thing (they even do use polling stealing CPU cycles in favor of faster I/O response) under Windows. Solaris/FreeBSD have ZFS with multi-level (RAM and SSD, something we're still missing...) ARC. That thing shines
Why is Starwind missing "multi-level ARC" (like ZFS)?

Or is this part of the upcoming version (the one you say is under heavy development)

awedio · Mon Sep 03, 2012 8:01 am

Is the new version in beta yet?

Mon Sep 03, 2012 8:02 am

Alpha... No public beta yet.

awedio wrote:Is the new version in beta yet?

awedio · Mon Sep 03, 2012 8:18 am

anton (staff) wrote:
Built-in NAS configurator allows you to create simple NAS (CIFS/SMB or NFS) failover cluster on top of StarWind so you'll be able to either
provide HA file share to your users or to keep VHD images and run Hyper-V from it (as what you wanted). But SoFS is preferred here as it allows to have some extra benefits (SMB Direct, pseudo-multipath based on DNS and so on).

I need to go back & look. I somehow missed the NAS configurator