I have a A1SRi-2758F , 32GB DDR, Samsung 845DC EVO MZ-7GE960 as local datastore right now, ConnectX-4 LX running ESXI 6.5 U1 and I created a freenas VM serving up two SanDisk SDSSDH31000G mirrored as my ESXI datastore for other ESXI machines. Performance is only hitting around ~150MB/sec so I started looking into things.
I came to the conclusion that the network interface is the bottleneck. I using a vmxnet3 adapter in Freenas and have already done a lot go tweaking like 4 RX/TX queues, network stack buffer optimizations and all I am gettin from the ESXI shell to Freenas running iPerf is around ~1.3Gbit/sec and maybe around ~1.3 to 3GBit/sec when testing with iPerf and -P 4 (4 threads).
I fired up a windows 10 machine and I am getting similar low speeds from ESXI to VM. When I do ESXI to ESXI, I get easily 8-9GBit/sec but I can't understand why the vmxnet3 interface is underperforming. When I look at top -SH on my freenas machine, I can see intr process taking a lot of CPU time.
Is this what I should expect from the C2758 architecture? Thanks for looking!
I came to the conclusion that the network interface is the bottleneck. I using a vmxnet3 adapter in Freenas and have already done a lot go tweaking like 4 RX/TX queues, network stack buffer optimizations and all I am gettin from the ESXI shell to Freenas running iPerf is around ~1.3Gbit/sec and maybe around ~1.3 to 3GBit/sec when testing with iPerf and -P 4 (4 threads).
I fired up a windows 10 machine and I am getting similar low speeds from ESXI to VM. When I do ESXI to ESXI, I get easily 8-9GBit/sec but I can't understand why the vmxnet3 interface is underperforming. When I look at top -SH on my freenas machine, I can see intr process taking a lot of CPU time.
Is this what I should expect from the C2758 architecture? Thanks for looking!