Hi Eli,<br><br>How did the CPU utilizations compare for the NAPI vs. no NAPI case? What are your thoughts on what bottleneck you are hitting?<br><br>Sorry to bother you ;)<br>thanks<br>harish<br><br><div><span class="gmail_quote">
On 9/21/06, <b class="gmail_sendername"><a href="mailto:eli@dev.mellanox.co.il">eli@dev.mellanox.co.il</a></b> <<a href="mailto:eli@dev.mellanox.co.il">eli@dev.mellanox.co.il</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
> Hi Eli,<br>><br>> Thanks for sharing the results with us. It is great to see the reduction<br>> in<br>> Interrupts. Could you please specify the netperf test specifications<br>> [message size; socket size]. Wondering what the numbers would be if we use
<br>> large socket and message sizes [128K & 64K respectively]. The reason for<br>> the<br>> request is to make sure we are not hitting any TCP related bottleneck<br>> while<br>> comparing NAPI vs. no NAPI cases. Please let me know what you think.
<br><br>I used large socket buffer sizes. Here is the command line I used. The<br>reult for the bandwidth is the some of all the connections.<br><br>netperf -H <a href="http://11.4.3.144">11.4.3.144</a> -l 600 -f M -p $port -- -s 200000,200000 -S
<br>200000,200000<br><br></blockquote></div><br>