[ovs-discuss] strange experiment result from ovs v1.9.3 in mininet

[Xiaoye Sun]

@ Han, I guess FLOW_MISS_MAX_BATCH should be the right place to change. For
my understanding, ovs userspace will batch all the packets it received and
the put packets from same flow into a "group", one group for each flow. So
I guess there are two levels of "batching".




On Thu, Jan 16, 2014 at 12:17 PM, Zhou, Han <hzhou8 at ebay.com> wrote:

> Hi Xiaoye,
>
> > Is there any way to disable this batching mechanism in the ovs kernel so
> that each upcall only contains one packet?
> Batching is a mechanism of the user space upcall handler rather than OVS
> kernel.
> To disable it you can change below macro in ofproto/ofproto-dpif-upcall.h:
> #define FLOW_MISS_MAX_BATCH 50 => 1
>
> However, I wonder how would batching affect your RTT result.
> On Wed, Jan 15, 2014 at 5:11 PM, Ben Pfaff <blp at nicira.com> wrote:
> > As the rate
> > increases, I guess that a batch contains more than one packet, and
> > it's likely that the batches mix up SYNs and SYN-ACKs, possibly
> > reducing the latency.
>
> @Ben, even if the batches mix up SYNs and SYN-ACKs, they are from different
> Flows. For SYN and SYN-ACK pairs of each single flow, they can never be
> mixed in the same batch, because SYN-ACK is not sent out until SYN is
> delivered
> to the receiver. So in my opinion, batching can improve the whole
> throughput but
> cannot explain the better STT. Did I miss anything here?
>
> @Xiaoye, maybe you can change FLOW_MISS_MAX_BATCH to different numbers to
> verify.
>
> Best regards,
> Han
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
>
>


-- 
Xiaoye (Steven) Sun, Ph.D. Student
Department of Electrical and Computer Engineering (ECE) & Department of
Computer Science (CS)
George R. Brown School of Engineering
Rice University, Houston, Texas, USA
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20140116/a9e8c1a9/attachment.html>