Re: What is the should be the expected latency of 10Gbit network connections

[Maged Mokhtar <mmokhtar@xxxxxxxxxxx>]

On 2018-01-22 08:39, Wido den Hollander wrote:

On 01/20/2018 02:02 PM, Marc Roos wrote:

  If I test my connections with sockperf via a 1Gbit switch I get around
25usec, when I test the 10Gbit connection via the switch I have around
12usec is that normal? Or should there be a differnce of 10x.

No, that's normal.

Tests with 8k ping packets over different links I did:

1GbE:  0.800ms
10GbE: 0.200ms
40GbE: 0.150ms

Wido

sockperf ping-pong

sockperf: Warmup stage (sending a few dummy messages)...
sockperf: Starting test...
sockperf: Test end (interrupted by timer)
sockperf: Test ended
sockperf: [Total Run] RunTime=10.100 sec; SentMessages=432875;
ReceivedMessages=432874
sockperf: ========= Printing statistics for Server No: 0
sockperf: [Valid Duration] RunTime=10.000 sec; SentMessages=428640;
ReceivedMessages=428640
sockperf: ====> avg-lat= 11.609 (std-dev=1.684)
sockperf: # dropped messages = 0; # duplicated messages = 0; #
out-of-order messages = 0
sockperf: Summary: Latency is 11.609 usec
sockperf: Total 428640 observations; each percentile contains 4286.40
observations
sockperf: ---> <MAX> observation =  856.944
sockperf: ---> percentile  99.99 =   39.789
sockperf: ---> percentile  99.90 =   20.550
sockperf: ---> percentile  99.50 =   17.094
sockperf: ---> percentile  99.00 =   15.578
sockperf: ---> percentile  95.00 =   12.838
sockperf: ---> percentile  90.00 =   12.299
sockperf: ---> percentile  75.00 =   11.844
sockperf: ---> percentile  50.00 =   11.409
sockperf: ---> percentile  25.00 =   11.124
sockperf: ---> <MIN> observation =    8.888

sockperf: Warmup stage (sending a few dummy messages)...
sockperf: Starting test...
sockperf: Test end (interrupted by timer)
sockperf: Test ended
sockperf: [Total Run] RunTime=1.100 sec; SentMessages=22065;
ReceivedMessages=22064
sockperf: ========= Printing statistics for Server No: 0
sockperf: [Valid Duration] RunTime=1.000 sec; SentMessages=20056;
ReceivedMessages=20056
sockperf: ====> avg-lat= 24.861 (std-dev=1.774)
sockperf: # dropped messages = 0; # duplicated messages = 0; #
out-of-order messages = 0
sockperf: Summary: Latency is 24.861 usec
sockperf: Total 20056 observations; each percentile contains 200.56
observations
sockperf: ---> <MAX> observation =   77.158
sockperf: ---> percentile  99.99 =   54.285
sockperf: ---> percentile  99.90 =   37.864
sockperf: ---> percentile  99.50 =   34.406
sockperf: ---> percentile  99.00 =   33.337
sockperf: ---> percentile  95.00 =   27.497
sockperf: ---> percentile  90.00 =   26.072
sockperf: ---> percentile  75.00 =   24.618
sockperf: ---> percentile  50.00 =   24.443
sockperf: ---> percentile  25.00 =   24.361
sockperf: ---> <MIN> observation =   16.746
[root@c01 sbin]# sockperf ping-pong -i 192.168.0.12 -p 5001 -t 10
sockperf: == version #2.6 ==
sockperf[CLIENT] send on:sockperf: using recvfrom() to block on
socket(s)








_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

I find the ping command with flood option handy to measure latency, gives stats min/max/average/std deviation

example:

ping  -c 100000 -f 10.0.1.12

Maged

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com