We are having performance issues with XEN as opposed to VMWARE. Pls take a look at the following cases (OVS - open vswitch, Bridge - native linux bridging)
Case 1 (no VM guest tagging): VM1 VNIC1 ---- (no tag) --- OVS ---(vlan X) --- PNIC1 ------(vlan X)--------- PNIC2 ---- (vlan X) ---- OVS --- (no tag) --- VNIC2 ---- VM2 Case 2 (no VM guest tagging): VM1 VNIC1 ---- (no tag) --- Bridge ------ PNIC1 ------(vlan X)--------- PNIC2 ---- (vlan X) ---- Bridge --- (no tag) --- VNIC2 ---- VM2 Case 3 (VM guest tagging): VM1 VNIC1 ---- (vlan X) --- OVS ---(vlan X) --- PNIC1 ------(vlan X)--------- PNIC2 ---- (vlan X) ---- OVS --- (vlan X) --- VNIC2 ---- VM2 Case 4 (VM guest tagging): VM1 VNIC1 ---- (vlan X) --- Bridge ---(vlan X) --- PNIC1 ------(vlan X)--------- PNIC2 ---- (vlan X) ---- Bridge --- (vlan X) --- VNIC2 ---- VM2 Case 4 does not work. So we moved to using OVS (open vswitch) because Case 3 works. But performance for Case 2 is much better than performance for Case 1. This is making us think about using bridging instead of open vswitch. Any reason why this is the case ? Has anyone done some benchmarking with xen+bridging and xen+open vswitch ? Pls point me to the results if they are available online. Also can someone pls point me to the data path architecture info for the open vswitch ? I understand that open vswitch can be deployed in user mode or kernel mode. We are mainly interested in kernel mode deployment for perf reasons. We are even considering building the whole thing pulling in the fixes ourselves as opposed to using the stuff that comes with xencenter 6.0 from citrix. Any pointers will be great. Thanks, -vijay
_______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
