Starwind N00B question

[altreed]

I have a test 2 node environment to create a small HA hyper-v cluster.

8 core AMD CPU @ 3.1GHz
16GB ram
120gb SSD boot
2x 1tb SAS in Raid0 through a RocketRaid controller
Onboard 1gb NIC
2x PCI 1gb Realtek Nic
Windows 2012
Starwind 6

I have been familiarising myself with the Starwind software, I have tcp4, jumbo frames, Full Duplex and optimised for iSCSI as per Starwind videos for the NIC’s

Onboard is LAN + heartbeat, 1 PCI Nic is Sync + HB and the other nic is for HV iSCSI.

I have tried a few builds to make a quorum and data stores. I have created a 50GB HA image and it was achieving 420mb/s through the sync as it was creating and syncing.

However, I had created one node with Raid0 and one with Raid1. As an experiment, I removed the replication partner on the Raid0, then converted the Raid1 to Raid0 and reconnected the replication partner, but it is taking forever to sync and I am seeing 14mb/s, as below. The behaviour at the moment looks like the nic is being throttled by something, but I don’t know where to look.



What can I do to allow the Sync link to deliver the 420MB/s performance?

Thanks

[Anatoly (staff)]

Hi!

First of all thank you for interest in StarWind!

As about your issue, I have few questions and I`d appreciate if you could answer them:
1) How did you measured above-mentioned 420 Mbps and 14 Mbps? Usually we expect to see the round figures here - 1 or 10 Gb/s, depending on NIC Card specification and number;
2) Is there any workload on the HA storage except synchronization process? The additional workload could obviously decrease the speed of the synchronization.
3) Can you confirm that RAID rebuild was successful on the time when the synchronization process was started?

Thank you!

[altreed]

Hi Anatoly,

In answer to your questions :-

1) How did you measured above-mentioned 420 Mbps and 14 Mbps? Usually we expect to see the round figures here - 1 or 10 Gb/s, depending on NIC Card specification and number;

I was watching the network performance tab on the task manager, as in the pic I was alerted to it as the sync progress bar was VERY SLOW in going up. I have resized the pic so it displays a bit better.

2) Is there any workload on the HA storage except synchronization process? The additional workload could obviously decrease the speed of the synchronization.

I have no HV's running accessing the HA images as yet, I am just building the bottom end, so to speak. I have got the cluster manager saying it is happy, HV's here we come

3) Can you confirm that RAID rebuild was successful on the time when the synchronization process was started?

The Raid setup was a quick init in the bios, it was 'instant' when I did the first Raid0 and built the HA images initially, seeing 420mbs as mentioned before. The only difference is that I created the HA image and got 420, rather than sync an existing node, the HAimage files are 50 GB, not TB. nb It was an Adaptec Raid card, not rocketraid.



When I started this project I made Raid 1 nodes and installed Windows etc, I was reading/watching one of the Starwind help resources and it introduced me to the concept of a 3 way raid 0 san and not use parity (faster) . So I thought I would try it and test the node failure/rebuild. As the sync seemed a fairly painful process I blitzed what I had set up and recreated the HA images (imagining that the Raid 1 was holding me back). As my experience of a 'throttled' sync I thought I would ask, in case my lack of knowledge/experience may have missed some vital config process.

Thanks
Dave

[altreed]

Hi,

Having synchronised over night and while I was at work, I just added the Hyper-v role to one node, the node rebooted as part of the install process and now the sync nic has variable connectivity, unable to reliably ping the other node. I have connected it through a dedicated 1g switch, but it have very intermittent connectivity. I am confused!

Right, I set one node pinging the other node over the sync channel. I rebooted the 'other node' and as it came back up the ping was reliable. I logged into the 'other node' and watched what happened. As soon as the starwind service started I got this ping behaviour :-



Then the behaviour is cycling through ping to general failure to request timed out, ad infinitum.

Thanks
Dave

[Anatoly (staff)]

That looks like something is wrong with the network part here for me.

Do you see any network related errors or warnings in the windows logs?