StarWind V5.8 Beta

Public beta (bugs, reports, suggestions, features and requests)

Moderators: anton (staff), art (staff), Max (staff), Anatoly (staff)

Post Reply
User avatar
Bohdan (staff)
Staff
Posts: 435
Joined: Wed May 23, 2007 12:58 pm

Wed Nov 23, 2011 3:58 pm

Dear Beta-Team members,

StarWind Software presents the new version of its main storage solution – StarWind iSCSi SAN 5.8 (beta version) - now enhanced by the powerful Backup technology included as a plugin. Today we are happy to provide our customers with the storage and backup software developed by the same vendor.

Backup plugin is built specifically for Hyper-V-based environments to provide fast backup and restore for Hyper-V virtual machines. It is very easy to set up and use; it makes backups just a simple task and allows fast restore of both virtual machines and individual files. Hyper-V Backup plugin is integrated with StarWind Single Management Console that enables managing backup and storage from the single window.

New features and improvements

Deduplication: Plugin has been reworked to support data journaling. Snapshots support added.

High availability: New version of plugin.
- Standard raw image files can be used for creating HA devices.
- MS iSCSI transport for sync channel has been substituted by new transport implementation.
- Several network interfaces can be used for synchronization and heartbeat channel.
- Heartbeat channel setting is required. It’s recommended to use client network as heartbeat channel. This channel is used for transferring service messages only, so it doesn’t produce any substantial network traffic.
- Switch partner feature. New partner node can be assigned to existing HA device node. Old partner must not be used after switching HA device to the new partner node. Client MPIO connections should be reconfigured to use path to the new partner instead of the old one.
- Automatic HA device synchronization after both nodes were down. If HA device can identify which node has the most recent and correct data, it brings this node online automatically and synchronizes the other node. If device cache mode is write-back and at least one node has not been turned off normally (for example, both nodes were down at the same time after power loss), HA device is not able to run autosync procedure.

RAM disk: serial Id generation algorithm changed. Now serial id is unique across different machines in the network. You can still add predefined IDs to configuration file for RAM devices.

SPTI: Buffer alignment parameter is used when checking device properties on service initialization.

Known issues
Deduplication: specifying custom snapshot name causes impossibility to create new device based on this snapshot.
HA: requires performance improvements

Please do not hesitate to contact us should you have any questions.


UDP:
StarWind 5.8 beta build has been updated:
http://starwindsoftware.com/tmplink/StarWindBeta.exe

New features and improvements

Core: Multi-LUN support.
HA: performance improvements.
HA: Fix for synchronization of HA after swith partner command
Deduplication: creation and mounting of deduplicated devices in background mode.
Fixes for snapshot processing in IBV and Deduplication devices.

INFO:
5.8 beta feedback, some questions/answers and feature requests can be found on the following page
http://www.starwindsoftware.com/forums/ ... t2644.html
Last edited by Bohdan (staff) on Tue Jan 03, 2012 8:23 am, edited 3 times in total.
User avatar
Aitor_Ibarra
Posts: 163
Joined: Wed Nov 05, 2008 1:22 pm
Location: London

Thu Nov 24, 2011 11:42 am

MS iSCSI transport for sync channel has been substituted by new transport implementation.
Is the MS iSCSI service still used for other functions, as the installer checks for it and won't proceed until it's running.

cheers,

Aitor
User avatar
Alex (staff)
Staff
Posts: 177
Joined: Sat Jun 26, 2004 8:49 am

Thu Nov 24, 2011 11:50 am

Aitor,
It has been retained as an optional transport for HA. It can be set up in HA plugin section of config file.
Best regards,
Alexey.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Fri Nov 25, 2011 10:48 pm

We're getting rid of Microsoft iSCSI stack as it's slow (our user-land implementation has smaller latency and better sustained transfer rate), buggy (incomplete transfer bugs I've reported back in 2006 or so are still there for example...) and unreliable (broken HA network connection should not bring whole machine to BSOD b/c of lost IRP somethere between storage and network stack) so it's kept for legacy purpose only. We'll remove mandatory check. Thank you for pointing!
Aitor_Ibarra wrote:
MS iSCSI transport for sync channel has been substituted by new transport implementation.
Is the MS iSCSI service still used for other functions, as the installer checks for it and won't proceed until it's running.

cheers,

Aitor
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
User avatar
Aitor_Ibarra
Posts: 163
Joined: Wed Nov 05, 2008 1:22 pm
Location: London

Sun Nov 27, 2011 2:37 pm

With HA targets, I know you say you need to optimise performance. I'm consistently getting lower write speed than read speed - is WB cache disabled, so if you set cache as WB currently, you actually get WT cache? I think WT cache MUST be working as I still get better speeds in an ATTO test than the underlying storage is capable of.

The only way I've managed to kill 5.8 so far is by making the classic mistake of leaving Widows Update on automatic, ending up with several instances rebooting at similar times. Even then, I was able to recover with no data loss because nothing was using the storage at the time. I've noticed that SW5.8 (perhaps earlier versions too, I never tried this before) can cope with the underlying storage disappearing on one node. E.g. when using another iSCSI target for the underlying storage of a 5.8 HA node, if you take the underlyind storage away, 5.8 will auto resync when it comes back. I wonder if this will work with a disappearing RAID set? Useful for anyone unfortunate enough to have WD Velociraptors with the 45 day bug!

As for MS iSCSI initiator, I've not really had a problem with performance over 10G. Perhaps at faster speeds it doesn't work well. I have had occasional BSODs on hyper-v nodes if both SW nodes become unavailable at the same time, which shouldn't happen, of course, only in the VMs using the storage.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Sun Nov 27, 2011 8:17 pm

This is not true as write is acknowledged only if second side sends ACK on it. We don't complete writes immediately if they are in LOCAL cache only (obvious reasons).

Yes, we've added basic journalling / transaction file system to dedupe (that's why image format is different from the version we had before) so it's very difficult to put it on the knees :)

Yes, should work fine the way you've told.

Peformance was not an issue - BSODs inside network / storage stack were...
Aitor_Ibarra wrote:With HA targets, I know you say you need to optimise performance. I'm consistently getting lower write speed than read speed - is WB cache disabled, so if you set cache as WB currently, you actually get WT cache? I think WT cache MUST be working as I still get better speeds in an ATTO test than the underlying storage is capable of.

The only way I've managed to kill 5.8 so far is by making the classic mistake of leaving Widows Update on automatic, ending up with several instances rebooting at similar times. Even then, I was able to recover with no data loss because nothing was using the storage at the time. I've noticed that SW5.8 (perhaps earlier versions too, I never tried this before) can cope with the underlying storage disappearing on one node. E.g. when using another iSCSI target for the underlying storage of a 5.8 HA node, if you take the underlyind storage away, 5.8 will auto resync when it comes back. I wonder if this will work with a disappearing RAID set? Useful for anyone unfortunate enough to have WD Velociraptors with the 45 day bug!

As for MS iSCSI initiator, I've not really had a problem with performance over 10G. Perhaps at faster speeds it doesn't work well. I have had occasional BSODs on hyper-v nodes if both SW nodes become unavailable at the same time, which shouldn't happen, of course, only in the VMs using the storage.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
User avatar
Aitor_Ibarra
Posts: 163
Joined: Wed Nov 05, 2008 1:22 pm
Location: London

Mon Nov 28, 2011 2:00 pm

I was just looking for a reason for the lower write performance... I can see how the ack from 2nd server's cache would cause a delay - and I will look to see if I have a bottleneck in the sync channel as well.

I have yet to unleash my storage-killer tests, but I think that my immediate impression is correct - of all the Starwind betas I've looked at, this is the closest to release quality I've seen. I haven't yet come across a functionality issue that would require a new build. It seems very reliable, and performance is very good considering the resources I'm using for testing it. So far, it seems that the main task before release is filling in the gaps in the help file.

I would have liked to have seen dedupe come out of "experimental" and also HA dedupe - but then I'm sure both of them are coming soon!

MS iSCSI and BSODs - are you finding BSODs definitely caused by MS iSCSI drivers, TCP/IP stack etc, and not by dodgy NIC drivers? I had a lot of problems with Hyper-V & early Intel drivers (later ones are much better) and as far as I know there are still problems with Broadcom teaming. I'm glad that Microsoft is finally going to support teaming in the OS with Windows 8.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Mon Nov 28, 2011 4:25 pm

1) Good! That was a plan :)

2) We're going to drop term "experimental" and provide HA + dedupe in V6. I'm not sure about release but you'll definitely see early beta in December 2011. We could release dedupe already but had to re-design it as it now supports transactions and consistancy points. Like ZFS and WAFL.

3) That's a long story and to make it short: MS iSCSI initiator was bringing *parent* machine to BSODs (BTW, Intel 10 GbE hardware...) and own user-land implementation does not.
Aitor_Ibarra wrote:I was just looking for a reason for the lower write performance... I can see how the ack from 2nd server's cache would cause a delay - and I will look to see if I have a bottleneck in the sync channel as well.

I have yet to unleash my storage-killer tests, but I think that my immediate impression is correct - of all the Starwind betas I've looked at, this is the closest to release quality I've seen. I haven't yet come across a functionality issue that would require a new build. It seems very reliable, and performance is very good considering the resources I'm using for testing it. So far, it seems that the main task before release is filling in the gaps in the help file.

I would have liked to have seen dedupe come out of "experimental" and also HA dedupe - but then I'm sure both of them are coming soon!

MS iSCSI and BSODs - are you finding BSODs definitely caused by MS iSCSI drivers, TCP/IP stack etc, and not by dodgy NIC drivers? I had a lot of problems with Hyper-V & early Intel drivers (later ones are much better) and as far as I know there are still problems with Broadcom teaming. I'm glad that Microsoft is finally going to support teaming in the OS with Windows 8.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
kmax
Posts: 47
Joined: Thu Nov 04, 2010 3:37 pm

Mon Nov 28, 2011 6:52 pm

Just curious if you read up on the next SMB versions in regards to clustering/fault tolerance and how Starwind fits in with it all.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Mon Nov 28, 2011 8:53 pm

We did not read we actually run this stuff in our lab :)

The answer lives somewhere in between with "It depends" and "I don't know". Generally speaking we do provide better level of service (dedupe is inline rather then offline and has better saving number + we have compression Microsoft don't so expect us save more disk space), our HA is active-active and covers multiple nodes instead of failover Microsoft has with a lonely active node, we do replicate deduped and compressed data rather then throwing RAW stuff into replication channel (VM replication in case of MS) so we're sutable for WAN and they need actually LAN and Microsoft's idea of "Share Nothing Live Migration" can be hardly used with anything except very few VMs - too many data should be copied, they don't do HA with SNLM and we do and so on and on and on... So it's only time who's going to show do we have enough of points to sell or people will accept lower level stuff but being entirely free.
kmax wrote:Just curious if you read up on the next SMB versions in regards to clustering/fault tolerance and how Starwind fits in with it all.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
robnicholson
Posts: 359
Joined: Thu Apr 14, 2011 3:12 pm

Thu Dec 01, 2011 11:08 am

We're getting rid of Microsoft iSCSI stack as it's slow
Is this just on the server or on the initiators?

Cheers, Rob.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Thu Dec 01, 2011 11:12 am

Initiators used for our internal work.
robnicholson wrote:
We're getting rid of Microsoft iSCSI stack as it's slow
Is this just on the server or on the initiators?

Cheers, Rob.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
robnicholson
Posts: 359
Joined: Thu Apr 14, 2011 3:12 pm

Thu Dec 01, 2011 11:21 am

So the installation on servers using the Microsoft iSCSI initiator is unchanged?

Cheers, Rob.
User avatar
anton (staff)
Site Admin
Posts: 4008
Joined: Fri Jun 18, 2004 12:03 am
Location: British Virgin Islands
Contact:

Thu Dec 01, 2011 11:22 am

For now yes.
robnicholson wrote:So the installation on servers using the Microsoft iSCSI initiator is unchanged?

Cheers, Rob.
Regards,
Anton Kolomyeytsev

Chief Technology Officer & Chief Architect, StarWind Software

Image
Post Reply