[ovs-discuss] Strange QoS/Latency problem with OpenvSwitch

[Chu Duc Minh]

I'm running an OpenStack-cluster, and use OpenvSwitch's VLAN to network
isolation (each teanant)

When I ping between VLAN (from instance A of tenant X to instance B of
tenant Y), i got a very strange problem:

# ping* -i 1 *10.2.1.26
PING 10.2.1.26 (10.2.1.26) 56(84) bytes of data.
64 bytes from 10.2.1.26: icmp_req=1 ttl=62 time=2.02 ms
64 bytes from 10.2.1.26: icmp_req=2 ttl=62 time=*13.5 ms*
64 bytes from 10.2.1.26: icmp_req=3 ttl=62 time=1.35 ms
64 bytes from 10.2.1.26: icmp_req=4 ttl=62 time=*19.8 ms*
64 bytes from 10.2.1.26: icmp_req=5 ttl=62 time=0.900 ms
64 bytes from 10.2.1.26: icmp_req=6 ttl=62 time=2.84 ms
64 bytes from 10.2.1.26: icmp_req=7 ttl=62 time=*42.5 ms*
64 bytes from 10.2.1.26: icmp_req=8 ttl=62 time=1.09 ms

# ping *-i 0.001* 10.2.1.26
...
64 bytes from 10.2.1.26: icmp_req=1783 ttl=62 time=0.479 ms
64 bytes from 10.2.1.26: icmp_req=1784 ttl=62 time=0.436 ms
64 bytes from 10.2.1.26: icmp_req=1785 ttl=62 time=0.576 ms
64 bytes from 10.2.1.26: icmp_req=1786 ttl=62 time=0.349 ms
64 bytes from 10.2.1.26: icmp_req=1787 ttl=62 time=0.475 ms
64 bytes from 10.2.1.26: icmp_req=1788 ttl=62 time=0.567 ms
64 bytes from 10.2.1.26: icmp_req=1789 ttl=62 time=0.519 ms
64 bytes from 10.2.1.26: icmp_req=1790 ttl=62 time=0.421 ms
64 bytes from 10.2.1.26: icmp_req=1791 ttl=62 time=0.668 ms
64 bytes from 10.2.1.26: icmp_req=1792 ttl=62 time=0.519 ms

# ping *-i 0.001 -s 1300* 10.2.1.26
# ping -i 0.001 -s 1300 10.2.1.25
PING 10.2.1.25 (10.2.1.25) 1300(1328) bytes of data.
1308 bytes from 10.2.1.25: icmp_req=1 ttl=126 time=122 ms
1308 bytes from 10.2.1.25: icmp_req=2 ttl=126 time=129 ms
1308 bytes from 10.2.1.25: icmp_req=3 ttl=126 time=126 ms
1308 bytes from 10.2.1.25: icmp_req=4 ttl=126 time=144 ms
1308 bytes from 10.2.1.25: icmp_req=5 ttl=126 time=18.3 ms
1308 bytes from 10.2.1.25: icmp_req=6 ttl=126 time=8.24 ms
1308 bytes from 10.2.1.25: icmp_req=7 ttl=126 time=0.801 ms
1308 bytes from 10.2.1.25: icmp_req=8 ttl=126 time=0.828 ms
1308 bytes from 10.2.1.25: icmp_req=9 ttl=126 time=1.31 ms
1308 bytes from 10.2.1.25: icmp_req=10 ttl=126 time=1.11 ms
1308 bytes from 10.2.1.25: icmp_req=11 ttl=126 time=0.867 ms
1308 bytes from 10.2.1.25: icmp_req=12 ttl=126 time=0.704 ms
1308 bytes from 10.2.1.25: icmp_req=13 ttl=126 time=48.5 ms
1308 bytes from 10.2.1.25: icmp_req=14 ttl=126 time=38.9 ms


I don't know why latency is smaller and *steady* when i flood ping with
small packets.
And why latency is "jitter" when i flood ping with very large packets.

And how can i configure it?

Thank you very much!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20131205/9c5e3f3c/attachment.html>