Hello!
Surroundings:
Compute nodes: DELL _R720
CPU: E5-2640v2 * 2 (32-core)
MEM: 16G * 24
Disk: 2 * 600G SAS (Raid1) + 6 * 4T (Raid5) SATA
NIC: 1G * 2 + 10G * 2 (Intel 82599EB 10-Gigabit SFI / SFP +
Network test is divided into two groups of test, a test is done HelloWorld, and
the other is PhoInfo do. When using PhoInfo test, the difference between
virtual and physical machines is not large, but when using HelloWorld test, the
difference is very obvious, the virtual machine is only about 1/4 of a physical
machine. Our analysis of the causes, because the HelloWorld page is very small,
the testing process is equivalent to generating a lot of small packets, PhpInfo
opposite page a lot, so that data packets generated relatively large. When
tested in the packet, the network bottleneck in the PPS, we have repeatedly
tested, the performance of the virtual machine software bridge can only reach
about 5wPPS, this time OpenVswitch has reached the limit, and ordinary physical
NIC determined to reach 200wPPS.
OVS really packet forwarding is much PPS? When IO-intensive data traffic onto
smaller, will not lead to OVS to hang? Packet forwarding delay would likely
increase?
thank you!
_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss
