Hi list,
I'm working under CentOS 6.4 + Openstack Havana + Openvswitch 1.11.0_8ce28d-1.el6ost.
I have a 10 Gb/s network link between the network node and the compute node.
Here is my Test case:
1. Start iperf server on the network node:
ip netns exec qdhcp-1314f7bb-9b52-4db8-a677-a751e52aad0e iperf -s
2. Boot an instance (name instance1) with flavor m1.tiny on compute node 1
3. Start a iperf client in instance1:
iperf -c $dhcp_device_ip
4. Get a bandwidth between the network node and the instances
It is about 8.77 Gb/s for GRE and 7.45 Gb/s for VXLAN.
Then I start more instances on the same compute node (compute node 1), and start more iperf client and server pairs between the network node and these instances.
ð Then I re-do the pervious test case : only one iperf client connect to iperf server
ð Here is the test result, the bandwidth is in a drop trend while the active instances number keep increases:
Bandwidth (Gb/s)
Active instance number GRE VXLAN
1 8.77 7.45
2 8.49 7.43
4 7.81 7.72
8 7.58 7.54
12 7.34 7.11
16 8.21 6.88
20 6.90 6.73
24 6.95 6.73
28 6.71 6.45
32 6.57 6.23
36 5.97 6.12
40 5.40 5.93
44 5.49 5.73
48 4.51 5.83
52 4.50 5.57
56 4.18 5.07
60 4.28 4.88
Anyone has the idea why this happen ???
Thanks.
-chen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20140207/1a11931f/attachment.html>