* [dpdk-users] OVS-DPDK performance raises when stress is running on the same core
@ 2018-04-04 15:00 Dawid Deja
0 siblings, 0 replies; only message in thread
From: Dawid Deja @ 2018-04-04 15:00 UTC (permalink / raw)
To: users
Hello,
I was testing how the OVS-DPDK will perform when putted into cgroup. To
my surprise, the performance raises once ovs-vswitchd process was able
to consume only 50% of CPU. Here's what I've done:
1. I have configured OVS to use dpdk and to run on core no 2;
2. I have created bridge with datapath_type=netdev, added a dpdk
interface to it and connected a VM to this bridge via vhostuser port;
3. I have created new cgroup, set the cpu.shares to 512 and added the
pid of ovs-vswitchd process to tasks;
At that point, I measured the throuhput from the VM, via the dpdk
interface to the external machine using iperf3:
iperf3 -c 192.168.0.3 -M 104
The bandwith was around 40Mb/s. After that:
4. I have started stress process to stress one core (# stress -c 1);
5. I have added the stress process to cgroup created in point 3;
6. Using taskset, I've put stress on same core as ovs-vswitchd;
Then I run the iperf3 again, with same command. The bandwith was around
100 Mb/s.Can this be explained somehow?
The tests were run on host with Atom CPU (Intel(R) Atom(TM)
CPU C2758 @ 2.40GHz).
Thanks,
Dawid Deja
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2018-04-04 15:00 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-04-04 15:00 [dpdk-users] OVS-DPDK performance raises when stress is running on the same core Dawid Deja
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).