UNSOLVED
AP
Ashley Poxon
3 Posts
0
19368
May 3rd, 2013 04:00
Throughput Measuring
Hi,
I'm after a bit of advice about how to measure network throughput between the hosts > switches > compellent controllers. There is a 1GB iSCSI, multi-pathed for the traffic between the hosts and the controllers. There is also a 2GB bonded network for the guest VMs.
We have approx. 50 VM guests, running on Citrix XenServer 6.0.2 (5 hosts), and I wanted to know how to measure the network throughput as I have a concern that there could be a bottleneck here, and Fibre Channel or 10GB iSCSI is something to consider but the costs is substantial. I wanted to be certain the bandwidth is saturated before making the move.
Any pointers where to start? I have the Compellent Enterprise Manager client software but not sure where to see anything relating to network throughput. The switches are HP ProCurve 2810-48G (2 of them for resiliency)
BVienneau
115 Posts
1150
0
Posted May 22nd, 2013 08:00
A great place to start is to go to the IO Usage tab in Enterprise Manager. From there you can view KB/Sec, IO/Sec, and Latency on all of the connectivity points the Compellent can see:
1. Server HBA/iSCSI connection
2. Controller HBA card
3. Disk Volume
The one thing it cannot see is things like the uplink ports on the network switches. You would need to use the tools provided by HP to see if uplink ports are being saturated (if the servers and Compellent controllers are not on the same switches)
Otherwise in Enterprise Manager I would look at the servers and see if you are maxing out the HBA's and would need more HBA's for throughput. If you are not maxing them out, then move on to the Compellent HBA ports and see if you are maxing out the bandwidth on those.
If you are pushing the bandwidth then moving from 1 GB to 10 GB could help and so could moving to FC but going 1 GB to 10 GB may be a cheaper investment in the long run.
The typical rough estimate for 1 GB iSCSI is bandwidth can theoretically max out at around 100 MB/s (120 MB/s but you need to account for overhead.)
If bandwidth doesn't seem to be an issue, then the IOPS might be the problem and that would require more disks.
As always, I tell my customers that Copilot support is an awesome resource and can help you narrow down bottlenecks whether it be the server, controllers, or disk so don't hesitate to contact them with help on this as well!
Hope this helps!