Is this correct behavior? Can someone explain?

[O1982]

Hi All,

Loving vSAN so far, great product (still can't believe there's a free version), but I have some quick questions.

Setup is 2 Node starwind vSAN setup for HA shared storage (CSV for Failover Cluster) on Win Svr 2012r2.

Can someone just confirm that this is expected behavior?

Question 1: With all VMs running on one node and VMs running from a CSV (on iSCSI from vSAN) that this node is the 'owner', in Task Manager I see alot of traffic on the iSCSI interface going to the other node. Why?

Question 2: When the node is accessing it's local storage should I see traffic on any of the NICs? I thought it would use the virtual loopback device? (which I cannot see except with Wireshark)

Question 3: I see very little traffic on the Sync interface/NIC. Is this normal? I would expect the traffic I see on the iSCSI NIC to actually be on the Sync NIC. I checked the Replication manager has the Sync NIC listed as the Sync Interface

Many Thanks for any help

Owen

[Serhi]

Hello O1982,

1. All data in HA devices must be the same. Thus, if your virtual machine is located on node 1, all data is immediately replicated to node 2. The synchronization process between nodes uses a synchronization channel, so it is the reason why you see traffic on the synchronization channel.
2. Look upper
3. Sync traffic depends on your production.

More information about how StarWind does work you can see at the link below:

https://www.starwindsoftware.com/best-p ... practices/

[O1982]

Hi Serhi,

Thanks for you answers, I have added info after your answers below:

Hello O1982,

1. All data in HA devices must be the same. Thus, if your virtual machine is located on node 1, all data is immediately replicated to node 2. The synchronization process between nodes uses a synchronization channel, so it is the reason why you see traffic on the synchronization channel. I only see little ~8KB spikes on the Sync Channel, this is when there are 2 TFS build servers running a build - I would expect some data to be written to itself via the loopback and to the other node on the Sync channel, but I do not see that.
2. Look upper
3. Sync traffic depends on your production.

More information about how StarWind does work you can see at the link below:

https://www.starwindsoftware.com/best-p ... practices/

I had seen and read the best practices before and I just checked again - I see no problems. Maybe I should simplify/reword my question:

Why do I see traffic on the iSCSI channel when all the required data is stored locally? The other node has no VMs that access data on the vSAN volumes.

[Serhi]

Hello again, O1982.

Could you explain to me where the management interface is in your circuit?

I checked StarWind White Papers and I found this:

https://www.starwindsoftware.com/resour ... 9M6QH.webp

1 heartbeat/iSCSI (see at this link: https://knowledgebase.starwindsoftware. ... planation/)
1 synchronization
1 management (and Heartbeat too)

[O1982]

Ok I think I finally figured it out - MPIO policy was set to Round Robin for some of the targets in the iSCSI control panel. I set all to 'Failover Only' and if the Target was the host I set ("Edit...") the 127.0.0.1 connection as Active and this automatically set the remote IP connection to "Standby". When I was checking this there were two targets where the loopback/127.0.0.1 was not automatically set as Active when the target was local. Strange.

I don't know if I have an uncommon setup but in the guides I have read it always seems to assume that you have multiple NIC going to the other server, where you would need the Round Robin policy. For single point to point connections to your other server I don't think this is needed or beneficial and could lead to a misconfiguration like mine.

Does anyone think this would cause any problems?

[Oleg(staff)]

Hi,
You answered your question
StarWind is active-active replication.
With loopback and partner ISCSI connections you can get better performance, for example, on reading operations. MPIO policy is distributing the load between loopback and partner connections.
That is expected behaviour to see traffic ISCSI connection, that is why it is in use.