[ovs-discuss] Fwd: question regarding openflow flow table

Haris Rotsos cr409 at cam.ac.uk
Thu Dec 16 16:11:16 UTC 2010


Dear Ben,

thanks a lot for your quick answer.

Following your answer, the insertion of exact match rules in the
classifier table doesn't have an impact on the switching performance
and is the default behaviour expected from openvswitch?

The linear search mode for the classifier table, was only a personal
impression.

On 16 December 2010 16:05, Ben Pfaff <blp at nicira.com> wrote:
> On Thu, Dec 16, 2010 at 8:00 AM, Haris Rotsos <cr409 at cam.ac.uk> wrote:
>> I noticed using ovs-ofctl that openvswitch has 2 tables : hash with
>> wildcard 0x000000 and classifier with wildcard 0x3fffff. My
>> understanding is that the first table is used for exact matches flows,
>> while the second is used for wilcarded flows. When I insert the
>> wildcard flows in the switch, I notice that they are placed in the
>> classifier table. Doesn't this has a performance penalty? The has hash
>> table uses a hash function with seek time constant, while the
>> classifier uses a linear lookup with linear seek time depending on the
>> number of flows in the table.
>
> If you're using Open vSwitch, then you are wrong that the classifier
> uses linear lookup.  It does not.
>
> If you have large numbers of wildcarded flows, recent versions of Open
> vSwitch (from the Git repository) will have better performance than
> older ones, because the classifier's performance has been improved.
> They also report only a single table, but this is really not relevant
> to performance.
>



-- 
Charalampos Rotsos
PhD student
The University of Cambridge
Computer Laboratory
William Gates Building
JJ Thomson Avenue
Cambridge
CB3 0FD

Phone: +44-(0) 1223 767032
Email: cr409 at cl.cam.ac.uk




More information about the discuss mailing list