VM Storage VMotion

[DavidMcKnight]

I think I know the answer to this, I just want to have someone verify it.

If I have two storage servers and I storage vMotion a VMWare VM from one datastore on storage server A to different datastore on storage server B, during the migration I can see some 6 gigabit worth of network traffic between the VMHost and either datastore (using windows resource Monitor) and several hundred megabytes worth of disk traffic between starwind and the .IMG file representing the datastore on either storage server. This works just fine. I can migrate a 100GB VM in 70ish seconds.

But if I storage vMotion a VM between two different datastores on the same storage server I see next to no network traffic and massive disk traffic. I can migrate that same 100GB VM in 50ish seconds.

The question, why is there no network traffic when migrating vms between datastores on the same storage server?

[Michael (staff)]

As far as I understand, you should not see the traffic on vMotion network because you migrating VMs to the same server.
In both cases, you should see the traffic on StarWind synchronization network.

[DavidMcKnight]

Okay, so here's the issue. The datastores in question are non-HA. So there's no synchronization involved. I can normally migrate a VM (storage or host) and the end user is unaware of it. But, when I migrated a VM between two datastores on the same storage server the VM in question became unresponsive and other VMs on the same datastores went into an alarmed state with high disk latency. Meanwhile inside of windows resource manager (on the storage server in question) I was seeing disk activity of 800MB/s reads from one datastore and some 900MB/s writes to the other datastore. This is three to four times the disk traffic speeds I see when migrating between two different storage servers which is great, but the time it takes to migrate is like 10% faster not the 300% - 400% I would expect from such high disk traffic.

So whatever voodoo VMware and Starwind are doing to allow one datastore to talk directly to another datastore (on the same server), I think you need to check to make sure that there are enough open resources to allow other non vMotion traffic to get though. In the end I had to cancel the migrating and then move the VM to a temp datastore on another server before moving it once more to the datastore I really wanted it to stay on.

[Michael (staff)]

Hello!
I think we need more information about your storage configuration and vMotion network speed to answer the question. Could you please provide us with it?

[DavidMcKnight]

I am decommissioning some older storage servers and bringing online some new ones. The old storages servers have:

Supermicro X8ST3 motherboard
Intel i7 CPU X 990 @ 3.47GHz
Intel X520-DA2 Dual 10Gb NIC
LSI MegaRAID SAS 9265‐8i RAID Controller
24 600GB 15K SAS2 Seagate Cheetahs
100GB C: drive (Server 2008R2, Starwind)
11TB D: drive (Starwind target storage)

New Servers:

Supermicro X10SRi-F Motherboard
Intel E5-1630v4 @ 3.7GHz
Intel X710-DA2 Daul 10Gb NIC
Avago 3108 MegaRAID RAID Controller
24 480GB Samsung PM863a SSD
100GB C: drive (Server 2016, Starwind)
9.6TB D: drive (Starwind target storage)

These servers are redundantly connected to my vCenter cluster with dual 10Gb Network. All vSphere hosts are also networked at dual 10Gb. On the vSphere hosts, the iSCSI datastores are set to round-robin. I created six 1.5TB iSCSI targets on the new server each one to hold maybe a dozen VMs ranging from 100GB – 150GB each.

Is this the info you wanted?

[Michael (staff)]

Hello David,
Thank you for provided information.
Please collect the StarWind logs from the problem storage server and log a support case here: https://www.starwindsoftware.com/support-form Please do not forget to specify the time and date when the issue occurred.
I would recommend using a script from the following KB article:https://knowledgebase.starwindsoftware. ... ogs-bat-2/