Dear Sirs,
Thank You for your replies.
Wes Felter : "Is there documentation about how to perform per-tenant QoS on a
physical
port in an overlay network like Neutron ML2? Since the
tenants are on
the integration bridge and the physical port is on a
different (or no)
bridge, it's not obvious how to do this. An example QoS
policy would be
weighted fair queuing where the weight is equal to the
amount of CPU/RAM
purchased by the tenant. "
Thank You for such a wonderful idea. It has given us a direction to look into.
We will definately dig deeper into it. And implement, if feasible.
I will surely update on the documentation.
Ben Flaff : "What do you see as the most important deficiency?"
Being quite the beginners in this stuff, we found some very basic deficiencies
like :
1. QoS implementations in ovs only support a simple rate limiting , rather
than intelligent QoS with a minimum bandwidth
guarantee. With rate limiting, each queue is assigned a specific amount
of bandwidth that cannot be exceeded. But what if
the queue is being under-utilized ?
2. Uncontrolled incoming traffic at the device, to a VM that is sharing a
network device, can severely impact the performance of
other VMs because the decision to drop an incoming packet is taken
after the device has received the packet. This could
potentially cause a denial of service attack on the VMs sharing the
device.
Sorry, if i sound stupid, but we are still in the literature survey phase. :)
Regards,
Monica. _______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss
