[ovs-discuss] Performance of short-lived TCP sessions

[McGarvey, Kevin]

That's good news.  I'll test it.

Kevin

On 2/3/14 3:30 PM, "Justin Pettit" <jpettit at nicira.com> wrote:

>On 2/3/14, 12:17 PM, "McGarvey, Kevin" <kmcGarvey at verisign.com> wrote:
>
>>The TCP workload of a global DNS server consists of TCP sessions, each
>>with a single request and a response that is typically under 2KB, and
>>there is a very large number of source IPs.  What level of performance,
>>expressed in terms of sessions per second,
>> would be expected from Open vSwitch with this type of workload?.
>
>It depends on the flow table.  Starting in OVS 1.11, if you had a
>configuration that doesn't require matching the source IP or ephemeral TCP
>port, it should be possible to cache the flow in the kernel, which would
>yield very numbers.
>
>--Justin
>
>
>