Hi Erik, Just sharing our experience. We recently ran similar network performance tests. Lower throughput in across nodes scenario is because of additional vxlan header processing. Our tests were using uperf and not iperf and other hardware specs will also matter so it will not be an apple-to-apple comparison to compare our results with yours.
On retransmissions, there were no restransmissions in our test results. I am not sure but there could be a system tuning/configuration issue on your setup. HTH. -Vikas Choudhary On Tue, Mar 28, 2017 at 4:27 AM, Roosemeyers Erik <erik roosemeyers arxus eu > wrote: > Hi all, > > > > I'm testing network connectivity (speed and reliability) on a OpenShift > cluster 3.1. > > > > I do this with 2 pods, each based on iperf3 Docker image. > > > > First pod starts iperf3 as server, other one as iperf3 client with: iperf3 > -c iperf3 -t 120 (iperf3 name of server pod). > > > > What I notice is high throughput but also big retransmission rates: > > eg. > > duration Transfer Bandwidth > Retransmissions (TCP) > > 0.00-120.00 sec 214 GBytes 15.3 Gbits/sec 170478 > > > > If the pods are on 2 different nodes, the retransmission rate is smaller > (as is throughput): > > > > duration Transfer Bandwidth > Retransmissions (TCP) > > 0.00-120.00 sec 22.6 GBytes 1.62 Gbits/sec 3627 > > > > What I was wondering - what causes this higher retransmission rates ? > (Mostly, it's a sign the are network problems but is it ?) > > > > (I did some other tests, like deploying the containers via the Docker > daemon onto 1 server. And even on a local laptop with only Docker daemon > installed - and I always saw these tcp retransmission rates...) > > > > > > Thanks in advance ! > > > > kind regards, > > Erik >
_______________________________________________ dev mailing list [email protected] http://lists.openshift.redhat.com/openshiftmm/listinfo/dev
