Re: Why dual traffic in 100Mbps envrionment is much less than 100M?

To: iperf-users --at-- dast.nlanr.net

Subject: Re: Why dual traffic in 100Mbps envrionment is much less than 100M?

From: "fu zhijun" <fuzhijun.door --at-- gmail.com>

Date: Fri, 28 Apr 2006 23:29:23 +0800

Content-type: multipart/alternative; boundary="----=_Part_3047_22892729.1146238163343"

Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:in-reply-to:mime-version:content-type:references; b=Mc7nVqXCwyB/JoXJ+jbEnc3H4FOqBGOCYVa+eM53sNIAC4Giob/r0K53W2RYa0FIWzKIEuYn3zpXsNJ9cjycwVPbMiH0cFxIz7uJQ0naeuscfJsEkBBmNbsunyEmH0xIOWsYCc9zFcivzVRHb94IfY8Ic/q3wUqe+7Vdz3ObQ8c=

In-reply-to: <008e01c66ad6$46d5f350$f939dd8c@mcs.anl.gov>

References: <d7139db60604280755m7b651bcbk13b080c8fa10a710@mail.gmail.com> <008e01c66ad6$46d5f350$f939dd8c@mcs.anl.gov>

Reply-to: iperf-users --at-- dast.nlanr.net

Sender: owner-iperf-users --at-- dast.nlanr.net

Thanks for your help.

I forgot to mention I have tuned this variable.

It is used in qdisc code in dev_queue_xmit(), enlarge this variable can avoid potential packet loss, which I have meet in my previous test. But this time there is almost no packet loss in my test, so it's because this limit.

Also, if this kind of packet loss happened, there is often a significant impact on the performance. But currently,there is no sudden performance drop, the performance just can not grow to 90M.

It is normal that packet loss and re-transmission would impact performance. But with almost no packet loss and little retransmission, and the sender cwnd has grown to a large enough value, why can't the performance reach peak in dual direction test?

It seems weird, and it actually happens......

2006/4/28, William E. Allcock <allcock --at-- mcs.anl.gov>:

A glue I found is in the case of no delay, when I set a big buffer for iperf, the NIC driver called "netif_stop_queue" many times, while with a small buffer for iperf, there is seldom. I added some debug info in NIC driver to find this.

Is this related to the problem I met? And how I can locate and solve this performance problem?

I cant be sure, but that sounds like you might be seeing what is called a "send stall", you are over-running the NIC queue. Try this (assuming you are usign eth0):

ifconfig eth0 txqueuelen 5000

and see if that makes any difference.

Thanks a lot in advance.