Re: Virtualization - network latency

Thanks Toby. From: "Toby Corkindale" <toby@dryft.net>
Here are the results of KVM and LXC based virtual machines on Ubuntu 14.04.. Both look a fair bit better than the results you're seeing.
--- LXC ping from VM host --- 10 packets transmitted, 10 received, 0% packet loss, time 9000ms rtt min/avg/max/mdev = 0.042/0.050/0.074/0.009 ms
That's in the same kernel so that should be better. Here are FreeBSD jail version 1 (shared network stack): 10 packets transmitted, 10 packets received, 0.0% packet loss round-trip min/avg/max/stddev = 0.016/0.019/0.025/0.003 ms .. and jail version 2 (VIMAGE/VNET, own network stack) 10 packets transmitted, 10 packets received, 0.0% packet loss round-trip min/avg/max/stddev = 0.236/0.280/0.321/0.030 ms results. Looks as there is significant overhead in the later.
--- KVM ping from VM host --- 10 packets transmitted, 10 received, 0% packet loss, time 9002ms rtt min/avg/max/mdev = 0.188/0.306/0.402/0.057 ms
Compared to Oracle VM(Xen)
10 packets transmitted, 10 received, 0% packet loss, time 9002ms rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms
It's nearly doubling the speed.. I wonder whether this is typical for KVM vs. Xen, and how VMWare ESXi is faring. I do not have a KVM or ESXi setup ready but will try. Thanks Peter

Hi,
On 5 Feb 2015, at 2:14 pm, Peter Ross <Petros.Listig@fdrive.com.au> wrote:
Compared to Oracle VM(Xen)
10 packets transmitted, 10 received, 0% packet loss, time 9002ms rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms
It's nearly doubling the speed..
Can you provide more information about the OVM setup? What guest type? What virtualisation model, i.e. HVM/PVM? Thanks, Avi

On Thu, 5 Feb 2015, Peter Ross wrote:
Thanks Toby.
From: "Toby Corkindale" <toby@dryft.net>
10 packets transmitted, 10 received, 0% packet loss, time 9002ms rtt min/avg/max/mdev = 0.554/0.605/0.646/0.037 ms
It's nearly doubling the speed..
I wonder whether this is typical for KVM vs. Xen, and how VMWare ESXi is faring.
At work, 0.5ms between hosts, 0.5ms within an ESXi host (but then again, with the nexus 5k and 7k switches and distributed virtual switches, doesn't all traffic end up going out the box and into the core switch and back anyway? Hold on, I'll just make sure I'm pinging the same vlan. Yep, still 0.5ms. But then again, why do you care about latency? I've not seen a non-HPC workload that cares about 1ms ping times. -- Tim Connors
participants (3)
-
Avi Miller
-
Peter Ross
-
Tim Connors