how tso (tcp segmentation offload) effect TCP bidirectional test sending throughput.

[Abu Rasheda]

On Fri, Apr 11, 2014 at 3:10 AM, Varun Sharma <vsdssd at gmail.com> wrote:

> Hi,
>
> if we off tso (tcp segmentation offload) on sending side ,  *TCP
> Bidirectional* test sending throughput decrease as compare to *TCP
> Unidirectional *test sending throughput .
>
> whereas if we on tso(tcp segmentation offload) on sending side , both testgive almost same
> sending throughput result.
>
> is anyone know how tso(tcp segmentation offload) effect TCP Bidirectional
> sending throughput ?
>
> i am using iperf-2.0.5 for this case.
>
> There are couple of things here.

1. Depending on the NIC, it could potentially have hardware offload in the
hardware. So offing TSO might not off it.
2. There is Linux kernel offload for receive, so there might be kernel
offload for send also
3. Different cards deal with offloading differently, with very different
result
4. If TSO is actually off, it might mean that eventually acks from remote
side are delayed filling its buffers

So this is very card, kernel version dependent.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20140411/46d8fe9a/attachment.html