I have a test environment consisting of a new Dell 1950. Two quad CPU's. 16Gb Ram. 300Gb HD (mirrored pair of SAS 300's, 15k).
Two VM's. One Server 2003 64-Bit. The other Server 2003 32-Bit. Fully patched and VMWare tools installed. ESX server 3.5. Also fully patched. Two internal Broadcom 1G NICs, Quad port Intel 1G card. Both VM's are on the Dell's internal hard drive.
If I connect the two VM's to a virtual switch with no NIC associated with it, file copy speeds between the two VM's runs at 200 - 300 Mbits/sec. Attaching a NIC and moving files back and forth from a physical server gives speeds closer to 200 Mbit/sec.
I've searched everywhere and about the only suggestion I can find is to force the virtual nics in the VM's to 1000-full. No difference.
Any ideas?
Darron...
I think by copying files locally you won't see more throughput .
How fast is your cloning (on local disks) ?
Im configured this kind of servers by one customer - but they are attached to san. By copying files from one host (vm on san) to ph. server I saw peaks of 500 Mb/sec - but the san storage has a 2GB cache.
On a vswitch there are no collisions so vm's can communicate faster internally.
You're probably running into hard drive performance constraints rather than a network bottleneck.
Rather use a proper network benchmarking utility such as iperf or netperf (especially iperf is quite easily obtainiable and free: http://dast.nlanr.net/Projects/Iperf/)
Be sure to use a Window size of at least 128K or larger for TCP tests. Note that UDP tests are limited to 10Mbps unless you explicitly state differently
Example: Run "iperf -s -w 128K" on the "server" side you want to test. Run "iperf -c
You're right. Iperf reported a throughput of 940Mbits/sec from the outside and 1.5Gbit/sec running VM to VM.
The network performance counter in Virtual Center must read in KBytes / sec and not KBits. Is there a way to change that?
Thanks,
Darron...
Unless you run your own queries against the SQL database, I don't think it's possible, no.
But then, does it really matter if it's Bits or Bytes? As long as you're comparing apples with apples things should be fine.