new HA install but slow performance

[dreesedatalink]

Let me start with i am new to StarWind. This is my fist install of StarWind but have setup other installs with datacore.

I am running a HA setup with to start two XEN 5.6 nodes. Both san nodes are running 5.4 StarWind. The san nodes have six 1GB nics in three teams (LAN, SYNC, SAN) the LAN team is jacked into a LAN switch, the SYNC team is crossover to other SAN node SYNC team, and the SAN team connects to a separate SAN switch, which will eventually be two separate switches.

Each xen server node has four 1GB NICS two go to the LAN switch and two go to the SAN switch. Each nic is configured with an IP of the corresponding subnet.

When I move a 6GB file from san1 to san2 over any of the teams I get roughly 50% NW utilization. When I move a 6GB file from SAN1 or SAN2 to a VM with storage hosted in StarWind over either LAN or SAN network I get roughly 2-3% NW utilization. When I move a 6 GB file from SAN1 or SAN2 to a VM with storage on the local xen server over either LAN or SAN network I get roughly 3-4% NW utilization. When I copy a whole VM using StarWind storage to a new VM using xen server local storage I get roughly 12.5% NW utilization.

I am not sure if my issue is with StarWind or with the xen server NIC config. But hopefully someone can point me in the right direction.
Thanks,

[anton (staff)]

Start with upgrading to 5.5 as it has HA performance dramatically boosted up. If you'd continue to have issues we'd start looking closer at your particular case. Thanks!

[dreesedatalink]

Update:
I have upgraded one node and waiting on the storage resync before i can upgrade the other node.

[anton (staff)]

Good! Please keep us updated about your progress. Thank you!

Update:
I have upgraded one node and waiting on the storage resync before i can upgrade the other node.

[dreesedatalink]

I just wanted to let everyone know. I have recreated the LUNs for half of the servers in my xen infrastructure using write back cache instead of write through cache and the performance has basically tripled. And server boot times have been cut in half. i am now seeing 50% nic usage (which would be the max on a 2 nic failover bond) and with using a large (2-3Gb) cache setting per lun file copies under 2GB are almost instant.

I initially went with write trough cache due to the "what if" both nodes went down hard. but this performance increase far out weighs my concerns.

I am once again singing the praises of starwind. This performance is right along with another client setup i have using datacore. so for the cost and simplicity for other techs to manage once initially configured starwind is now my goto.

Thanks!

[anton (staff)]

This is exactly what we see in our lab and on major part of our customers sites as well. Thank you very much for detailed clarification!

I just wanted to let everyone know. I have recreated the LUNs for half of the servers in my xen infrastructure using write back cache instead of write through cache and the performance has basically tripled. And server boot times have been cut in half. i am now seeing 50% nic usage (which would be the max on a 2 nic failover bond) and with using a large (2-3Gb) cache setting per lun file copies under 2GB are almost instant.

I initially went with write trough cache due to the "what if" both nodes went down hard. but this performance increase far out weighs my concerns.

I am once again singing the praises of starwind. This performance is right along with another client setup i have using datacore. so for the cost and simplicity for other techs to manage once initially configured starwind is now my goto.

Thanks!