<div dir="ltr">On Fri, Apr 11, 2014 at 3:10 AM, Varun Sharma <span dir="ltr"><<a href="mailto:vsdssd@gmail.com" target="_blank">vsdssd@gmail.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div><div><div><div>Hi,<br></div><br></div>if we off tso (<span>tcp</span> segmentation offload) on <span>sending</span> side , <b>T<span>CP</span> <span><span>Bidirectional</span></span></b> <span><span>test</span></span> <span><span>sending</span></span> throughput <span><span>decrease</span></span> as compare to <b><span><span>TCP</span></span> Unidirectional </b><span><span>test</span></span> <span><span>sending</span></span> throughput <span></span>.<br>
<br></div>whereas if we on tso(<span>tcp</span> segmentation offload) on <span>sending</span> side , both <span>test</span> give almost same <span>sending</span> throughput result.<br>
<br></div>is anyone know how tso(<span>tcp</span> segmentation offload) effect T<span>CP</span> <span><span>Bidirectional</span></span><b> <span></span></b><span>sending</span> throughput ? <br>
<div><br></div><div>i am using iperf-2.0.5 for this case.</div></div>
<br></blockquote><div style>There are couple of things here. </div><div style><br></div><div style>1. Depending on the NIC, it could potentially have hardware offload in the hardware. So offing TSO might not off it.</div>
<div style>2. There is Linux kernel offload for receive, so there might be kernel offload for send also</div><div style>3. Different cards deal with offloading differently, with very different result</div><div style>4. If TSO is actually off, it might mean that eventually acks from remote side are delayed filling its buffers</div>
<div style><br></div><div style>So this is very card, kernel version dependent.</div></div></div></div>