StarWind Software

Hello, currently I'm trying to deploy VSAN Starwind Free for Failover Clustering 2 Windows Server with 2 Witness Node in different site. The placement on each site is 1 Windows Server and 1 Witness Node. And each NIC have it's own subnet like what's recommended by Starwind, and have different multiple physical connection for redundancy in sync channel and witness/heartbeat channel. The network speed is pretty limited, currently I only have around 150-300 Mbps for the network transfer between site. The idea for the witness is I use DFS Replication between the 2 Witness Node to setup SMB File Share to be used by VSAN as the SMB Witness path.

So, I've already done my setup and it's working nearly perfect. I'm using Failure Sync Journal, Write-Through cache for L1 & L2 cache, and also using Node Majority with SMB Witness for the failover strategy. So my experience with this setup is it's pretty robust, even when all network connection is down (even the different physical ones). After I bring it all up it always give me a Fast Synchronization, although maybe it still takes around 3-6 hours for the sync of 3 TB of image.

So my question is :
1. Is it normal when the HAImage is already synchronized and all sync channel lost connection, the HAImage doesn't auto sync? (Even though the witness channel is still up)
2. Do you have any recommendation for my setup? (Beside increasing my network speed of course)
3. When in Fast Synchronization, when one of my sync channel is down (The other sync channel is still up). Is it normal for the Synchronization to fail and Partner Image changes to Not Synchronize state?
4. Is it possible to change the synchronization journal type with Powershell CLI? If so, how do I change it?

That's all for my question, and thank you for all the Documentation and Forum Discussion you guys provide. It actually helps me a lot as I'm pretty new in this HA/Failover business. 10/10 would recommend my boss to buy the enterprise license

Welcome to StarWind Forum, and thanks for your kind words.
Having a Witness and a data node in the same location makes data unreachable when that side goes down, as the only data node will become "isolated" and VSAN will not publish HA storage from it.
As a side note, you cannot set up a HA device with 2 witness nodes.

Also, if the switches are redundant in each location, then you can use heartbeat.

The network speed is pretty limited, currently I only have around 150-300 Mbps for the network transfer between site.

We recommend at least 1 GBE. Also, make sure that the latency will be below 5 ms.

1. Is it normal when the HAImage is already synchronized and all sync channel lost connection, the HAImage doesn't auto sync? (Even though the witness channel is still up)

Yes.

2. Do you have any recommendation for my setup? (Beside increasing my network speed of course)

Remove L2 and L1 caches.
Make sure that the witness node/share is in an independent location.

3. When in Fast Synchronization, when one of my sync channel is down (The other sync channel is still up). Is it normal for the Synchronization to fail and Partner Image changes to Not Synchronize state?

Yes, they work in round-robin, and when the replication link drops, it is possible that the replication drops too.

4. Is it possible to change the synchronization journal type with Powershell CLI? If so, how do I change it?

No, only through header reconfigurations. I have submitted the feature request, though. Thanks for making our product better!

Good luck with your project!!

Hey, thank you for the fast response. Truly appreciate your recommendations, I will try to implement it. But could you elaborate on why removing the cache is actually better than having one or two?

You are welcome.
Write-back caching makes the systems touchy. Full synchronization and increased risks of a mutual not synchronized state that does not solve itself after power outages.
SSD caching is not highly available, which might result in data inconsistencies sometimes.

I see. Also, another question, you said when using multiple channels/interfaces they work in round-robin. Is it still possible to make the sync process redundant? So even when one sync interface/channel is down, the sync process is not interrupted and the replication progress not loss.

Yes, you can make it redundant, yet the links will be switching only following the round-robin strategy.
Under normal circumstances, a redundant sync link blip does not cause much drama. While full/fast sync is running, it does so as both links are utilized heavily at that moment.
Sync progress loss is another story. I filed a request a while ago about retaining sync progress so that it does not restart from scratch; it traveled to the road map, and we plan it down the road.
Sadly, at this moment, there's nothing we can do to make it happier

I see, pretty interesting. Hope this feature as you said will go to production soon. Really looking forward to it, for now I will play and see what VSAN Starwind will behave based on your recommendation earlier. Thank you so much for your responses, yaroslav. Have a nice day!

You are welcome
For POC, Trial works better, IMO, as you get tech assistance and GUI access. Also, it is a nice chance to learn more about our products in general.
There's a catch, though: you can't switch Free from Tiral.