Re: Reason to drop packets


- more information is necessary

The two systems are interconnected point-to-point via ethernrt wire  (
100Mbit /s  ), the CPU utilization  I see during the test is  ~ 90 %  (
using "top" command or system monitor ).
I start dropping packets when UDP bandwidt > 9Mbit / s.


- questions

You wrote : "using larger buffers may help this since there will be less
overhead", but what do you mean by buffer ? UDP buffer size or UDP packet
size ?
How can I know the lost packets number on outgoing interface queues ?


Thanks



----- Original Message -----
From: "Jon Dugan" <jdugan --at-- ncsa.uiuc.edu>
To: <iperf-users --at-- dast.nlanr.net>
Sent: Monday, February 23, 2004 10:01 PM
Subject: Re: Reason to drop packets


> On Mon, Feb 23, 2004 at 06:46:30PM +0100, onorati_m --at-- virgilio.it wrote:
> > Hi all!
> >
> > Could someone explain why Iperf really starts dropping packets?
> >
> > I mean, selecting the UDP mode, setting a UDP bandwidth value and the
> > number of bytes to be transmitted, why does Iperf show losses for higher
> > values of the UDP bandwdith rate?
> >
> > Is it a matter of CPU load or something related with the buffer sizes?
>
>   Iperf using UDP doesn't drop packets per se, rather it will blindly
transmit
>   at the specified rate and if there is contention for resources somewhere
in
>   the system the data will be lost.  Often it is CPU utilization that is
the
>   culprit.  What is the CPU utilization you see during the test?  Using
larger
>   buffers may help this since there will be less overhead.
>
>   The other common cause of lost data with a UDP test is loss somewhere in
the
>   network.  How are the two systems interconnected?
>
> Jon
> --
> Jon Dugan             |  Senior Network Engineer, NCSA Network Research
> jdugan --at-- ncsa.uiuc.edu  |  269 CAB, 605 E Springfield, Champaign, IL 61820
> 217-244-7715          |  http://www.ncsa.uiuc.edu/~jdugan/
>



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index