Starwind HA Cluster + HyperV

starneo · Tue Feb 08, 2011 11:27 am

I will send yout the crshdump from node B (nodeA never crashed) and the logfiles from both nodes. I can not find any logfiles on my management server (starwind console) and there is no dump, this server never crashed, too.
But at first I have to compress 2 Gb to a size I can send.

Tue Feb 08, 2011 11:29 am

starneo wrote:I will send yout the crshdump from node B (nodeA never crashed) and the logfiles from both nodes. I can not find any logfiles on my management server (starwind console) and there is no dump, this server never crashed, too.
But at first I have to compress 2 Gb to a size I can send.

If they refuse to zip to a small size -just put them on any filehosting service

Tue Feb 08, 2011 3:25 pm

...and if you don't have one we'll be happy to provide our ftp

Max (staff) wrote:
starneo wrote:I will send yout the crshdump from node B (nodeA never crashed) and the logfiles from both nodes. I can not find any logfiles on my management server (starwind console) and there is no dump, this server never crashed, too.
But at first I have to compress 2 Gb to a size I can send.
If they refuse to zip to a small size -just put them on any filehosting service

starneo · Tue Feb 08, 2011 3:26 pm

I uploaded it to our homepage and it has already been downloaded.

Tue Feb 08, 2011 3:28 pm

Excellent! So we're under way.

starneo wrote:I uploaded it to our homepage and it has already been downloaded.

starneo · Wed Feb 09, 2011 8:50 am

I removed all targets and create new ones.
This time I user Starwind node A as primary only and Starwind node B as partner/secondary only.
I will do some testing with this configuration.
Now I have to configure all iscsi connection with mpio on the 3 servers, which means 3 server * 2 nodes * 5 targets * 4NICS (MPIO) = 120 connections over all.
And there are 4 more server to come ...
So my question is, is there a better tool to manage this, especially on Windows 2k8R2 ServerCore?
I can do it manualy by clicking through the GUI, but I think it is not a good way, because I can easily do some misstakes.
A tool with a config file or a good script would be nice

Wed Feb 09, 2011 9:21 am

For now it's entirely "Fire-and-Forget" scenario. But we'll be adding scripting to StarWind pretty soon.

starneo wrote:I removed all targets and create new ones.
This time I user Starwind node A as primary only and Starwind node B as partner/secondary only.
I will do some testing with this configuration.
Now I have to configure all iscsi connection with mpio on the 3 servers, which means 3 server * 2 nodes * 5 targets * 4NICS (MPIO) = 120 connections over all.
And there are 4 more server to come ...
So my question is, is there a better tool to manage this, especially on Windows 2k8R2 ServerCore?
I can do it manualy by clicking through the GUI, but I think it is not a good way, because I can easily do some misstakes.
A tool with a config file or a good script would be nice

starneo · Fri Feb 11, 2011 10:08 am

Yesterday I finished my new configuration.
This time I have
-6 targets over all
-6 primary on node A (4x4Gb cahce + 1x2GB cache +1x64MB cache)
-6 secondary (partner) on node B (4x4Gb cahce + 1x2GB cache +1x64MB cache)

Everything was fine so far, in my first test I was not able to reproduce the BSOD/starwind service error descibed in the former posts.
But I found a new problem ...

Yesterdy in the evening I shuted the HyperV Cluster down and Starwind Node B ( Node A was running over the night). Today I started Starwind Node B and simultaiously the 3 HyperV Server (Cluster). The HyperV Servers got a connection, everything was good so far. Then on Starwind node B the starwind service came up and here was the problem.
The node started a "Full Sync" and not a Fast Sync ... but thats not all, the Sync speed was horrible slow (2-8% of 10Gbit, should be ~60%). So I restarted starwind node B.
Okay same problem, full sync - this time the speed was bette but not constant. The speed was between 10%-50%, jumping up and down ( to be clear, the speed is at 10% but jumping every few seconds to 50%, so overall 20%). Normaly when doing a full sync the speed ist constant at 50%-65%. The other thing thats irritating, on Starwind node A the RAM load is 4.4GB and on the fresh rebootet node B 19,7GB.

So my question is, why does Starwind do a full Sync?

(I want to go live at 01.03.2011 with the whole system and have to do much more ... and now I lost a whole day becouse of the syncing ( performance during sync isn't good, but that I do understand!))

Fri Feb 11, 2011 10:27 am

The fullsync was launched because of the fastsync limitations.
the amount of data which can be transfered during a fast synchronization session is equal to total disk size/65536=max number of write commands included in the Fastsync log, in case the difference between the nodes can't fit in this number - a full sync occurs. Another 2 triggers for the full sync are sync channel loss and sync abort.

starneo · Fri Feb 11, 2011 10:34 am

Oh, so that means, runnning a Full HyperV Cluster (100 - 150 VMs) with the Starwind HA Set and rebooting one Node for installing Windos Updates/ServicePack or a Starwind Update will result to 99% in a full sync? (Rebooting takes arround 3 mins for Windows + 10-15 mins for starwind service)

Thats good to know, now I have to overthink my plans for the maintenance.

Do you have any Idea for the speed issue?

Fri Feb 11, 2011 1:58 pm

1) When difference between nodes is at some high watermark (you can change it yourself BTW) it makes sense to run full sync b/c it's actually faster. Doing linear sequential write with 64KB - 256KB packets is faster compared to random I/O with small chunks over the write bitmap (what fast sync does by it's nature).

2) In your case fast was escalated to full b/c you've been doing sync already. Failure during sync (fast or full - does not matter) makes StarWind run full sync with the next attempt.

starneo wrote:Oh, so that means, runnning a Full HyperV Cluster (100 - 150 VMs) with the Starwind HA Set and rebooting one Node for installing Windos Updates/ServicePack or a Starwind Update will result to 99% in a full sync? (Rebooting takes arround 3 mins for Windows + 10-15 mins for starwind service)

Thats good to know, now I have to overthink my plans for the maintenance.

Do you have any Idea for the speed issue?

starneo · Fri Feb 11, 2011 2:55 pm

okay, understood.
So I can change the percentage when starwind will do a full sync from 0.01% to lets say 0.1% or 1%? Can you tell me where?

Fri Feb 11, 2011 3:11 pm

In production version you cannot change this. I'm really sorry for confusion ((

starneo wrote:okay, understood.
So I can change the percentage when starwind will do a full sync from 0.01% to lets say 0.1% or 1%? Can you tell me where?

Fri Feb 11, 2011 3:16 pm

I've got good news for you
We have just released 5.6 build and it may help you to fix the issues.
Please update the second node (the one which is failed) http://www.starwindsoftware.com/downloads/
The targets will now synchronize one by one and bit more fast.
Please keep us posted and don't hesitate to ask any questions.

starneo · Fri Feb 11, 2011 4:38 pm

great! *deleted* I'll upgrade both nodes now