StarWind V5.9/6.0 Beta

Vavaga · Mon Jun 11, 2012 6:09 am

Hello, i cannot add existing devaice. On stage interfaces I'w got error "Please, select interfaces for a sync channel", but there is no sync chanel option to add (see scrins)

: Specify Interfaces.jpg (87.11 KiB) Viewed 83986 times

: selectinterfaceerror.jpg (28.14 KiB) Viewed 83988 times

Mon Jun 11, 2012 6:34 am

Hi
Is it latest (20120426) beta build?
http://starwindsoftware.com/tmplink/StarWindBeta.exe

Vavaga · Mon Jun 11, 2012 8:34 am

Bohdan (staff) wrote:Hi
Is it latest (20120426) beta build?
http://starwindsoftware.com/tmplink/StarWindBeta.exe

Yes, I have just download it on friday 08 jun 2012 from link above.

Mon Jun 11, 2012 9:01 am

Please make sure that all HA nodes has the latest 5.9 StarWind service. Also please make sure that you are using proper StarWind Management Console.
In case you are managing StarWind servers remotely, you should use the management console 5.9.2025 (it is a part of the latest beta build).

Vavaga · Mon Jun 11, 2012 3:51 pm

Bohdan (staff) wrote:Please make sure that all HA nodes has the latest 5.9 StarWind service. Also please make sure that you are using proper StarWind Management Console.
In case you are managing StarWind servers remotely, you should use the management console 5.9.2025 (it is a part of the latest beta build).

Yes all nodes and console use latest beta build. This problem hapend if one HA target on node was delited and you try to add again delited partner image disk to existing HA.

Tue Jun 12, 2012 7:59 am

Got it! Thanks.
This problem is included in known issues list ("HA: Switch partner functionality is not implemented yet"), that's why I was a little confused.
It will be implemented in the next beta build. I'll send the notifications as soon as it is ready for download.

Vavaga · Tue Jun 12, 2012 9:39 am

Hello! Why network speed of sunchronization proccess depend of HA device cache parameters? For sample if set create HA with no cashe parameter, the synchronization network speed will keep 20% of 1Gbit net(firs graf on scrinshot). If set create HA with write-back cahe 64Mb maximum cashe size, the network will jump on 50% for a second and then down on 5% of 1Gbit net(second graf on the same scrinshot). If set create HA with write-back cahe 1024Mb maximum cashe size the network will keep 75% till cashe will full and then jump down on 5% of 1Gbit net (third graffik on the same screen).

: netspeed.jpg (201.71 KiB) Viewed 83881 times

: createHAwrbackcache64Mb.jpg (57.15 KiB) Viewed 83878 times

: createHAwrbackcache1024Mb.jpg (57.42 KiB) Viewed 83881 times

Tue Jun 12, 2012 10:07 am

In most cases fast sync causes random disk access, while full sync - sequential.
Synchronization speed depends on hard disk speed, but not only the network interface speed. It varies depending on the access block size (4K, 8K, ... 64K, 128K ...). Cache synchronization is fast, because the underlaying storage (RAM) is fast. It causes the better network utilization (peak on the graph).
BTW, are those HA nodes physical or virtual machines?

Vavaga · Tue Jun 12, 2012 11:02 am

Bohdan (staff) wrote:In most cases fast sync causes random disk access, while full sync - sequential.
Synchronization speed depends on hard disk speed, but not only the network interface speed. It varies depending on the access block size (4K, 8K, ... 64K, 128K ...). Cache synchronization is fast, because the underlaying storage (RAM) is fast. It causes the better network utilization (peak on the graph).
BTW, are those HA nodes physical or virtual machines?

HA nodes are virtual. I use two server dl380g7 with esxi 5. Starwind instaled on both hosts on w2008r2 virtual servers. Managment throught vcenter 5. But problem that cashe synchronization is fast only for period when cash is not full. As soon as 1024mb cashe filled the speed jump down from 75% to 5% but cashe is not cleaning down after 5 seconds.

: cpu.jpg (197.21 KiB) Viewed 83864 times

Tue Jun 12, 2012 12:35 pm

Do you have some other disk activity on the physical disks of these ESX servers?
It looks like you are getting real disk performance when the cache is full. Low disk performance can be caused by some other activities on the hard disk that is not related to StarWind virtual device. In this case hard disk is getting random access instead of sequential access.
You can check the disk performance: create StarWind ImageFile and try to copy some large file to this device. Or check it with IOMeter.

Cache is not cleaning down after 5 seconds.

Cache is not intended to be cleaned. 5 seconds is a time that service will wait before flushing new data. It the data is accessed again during this time, it will not be flushed to disk 5 seconds more. When the data is flushed, it remains in memory to serve as read cache.

Vavaga · Wed Jun 13, 2012 8:48 am

Alex (staff) wrote:Do you have some other disk activity on the physical disks of these ESX servers?
It looks like you are getting real disk performance when the cache is full. Low disk performance can be caused by some other activities on the hard disk that is not related to StarWind virtual device. In this case hard disk is getting random access instead of sequential access.
You can check the disk performance: create StarWind ImageFile and try to copy some large file to this device. Or check it with IOMeter.

Cache is not cleaning down after 5 seconds.
Cache is not intended to be cleaned. 5 seconds is a time that service will wait before flushing new data. It the data is accessed again during this time, it will not be flushed to disk 5 seconds more. When the data is flushed, it remains in memory to serve as read cache.

Seems the problem with esxi. If I copy file to LUN of starwind image created with cashe option from vcenter, speed is about 150mbit/s(frist graffick on screenshot). If I copy file directly from one server to another through network speed shows 900 mbit(second graffick on screenshot) . Servers are virtual and placed on different esxi hosts.

: img&directcopy.gif (52.27 KiB) Viewed 83754 times

Vavaga · Wed Jun 13, 2012 9:07 am

Could you please give link on document where can i reed how HA storage system should start work? after power down on both hosts? Becourse now after power down? HA storage do not start automaticly synchronisation and not starts synchronisation manualy

: synccannotstart.gif (20.5 KiB) Viewed 83752 times

Only command "Mark as Synchronized" back system on work. Is it normal?

Wed Jun 13, 2012 9:31 am

Yes it is normal if you are using write-back cache and pressed reset button or performed force shutdown. This case the data in RAM are lost.
In case of WT or no-caching configuration synchronization should start.

Vavaga · Wed Jun 13, 2012 1:05 pm

Bohdan (staff) wrote:Yes it is normal if you are using write-back cache and pressed reset button or performed force shutdown. This case the data in RAM are lost.
In case of WT or no-caching configuration synchronization should start.

Ok I had try. The results below:
Write-through cashing configuration - automatik sunchronisation did not start after power down on both hosts, and did not start manualy. Only "Mark as synchronizaded" option helps to start HA storage.
No-cashing configuration - automatik sunchronisation did not start after power down on both hosts, and did not start manualy. Only "Mark as synchronizaded" option helps to start HA storage.
It means that if power will be down and up after some time, all virtual mashines and their services working on HA storage wil be down. Is it problem only of of 5.9 beta version? Also in 5.8 product there was option "Automtikly star sync after fault" why tyere is no such option in 5.9?

Wed Jun 13, 2012 3:31 pm

Thanks!
Auto restore is disabled in current 5.9 beta build by Dev team for debug purposes.
I'm sorry for misleading.

Also in 5.8 product there was option "Automtikly star sync after fault" why tyere is no such option in 5.9?

Node auto restore will be the default behavior in 5.9 release.