[Mujahid Syari <Syari.Mujahid@xxxxxxxxx>]

Thanks Johannes for your input.

In our test, we use our home grown product to send packets (machine <->
machine cable connection). Our destination machine uses the 3c59x driver
(3COM), it seems that we just lost small number of packets even when go up
to 7.0 Mbps. but one thing is sure that as soon as we reach 0% of CPU idle
time, that is the time when we lost packet. We have tried it with using
different machine and different OS (Solaris and Linux) to test which one
performs better. It seems to us Solaris (Sparc) is better compare to Linux
(both of them are out of box). I still don't understand why Linux
performance (Mandrake 8.1) is less compare to Solaris. This is the reason
why I send this message. It seems to me there must be a way to tweak Linux
to make it faster then Solaris. I have seen a paper concerning NAPI (ftp
robur.slu.se:/pub/Linux/net-development/NAPI) which apparently gives better
performance on Linux, but I have no time to try it yet.

Maybe I need a strip down version of Linux, with Pentium 2.4 Ghz, 4 CPU,
e1000 network card, Flashdisk, 2 Giga of RAM. But I don't know if this is
necessary.

Regards,

Syari

> -----Original Message-----
> From:	Johannes Faerber [SMTP:faerber@xxxxxxxxxxxxxxxxxxxx]
> Sent:	mardi, 7. mai 2002 11:23
> To:	Ethereal-users@xxxxxxxxxxxx
> Cc:	Mujahid Syari
> Subject:	Re: FW: [Ethereal-users] What is the peak limit for Ethereal
> ?
> 
> Besides CPU, Harddisk and Bus the capture performance of ethereal
> or tcpdump depends heavily on the network card & driver and wether you 
> capture full packets or only the packet header (first 68 Bytes). 
> 
> In a test measurement of packet headers with a peak rate of 12000
> packets/s 
> we have seen packet loss of 6*10^-5. Two similar machines showed 8*10-6.
> 
> We tested "Intel Ether Express Pro 100" against "DECchip 21140" (Tulip) on
> an IDE
> PIII 550, 384 RAM. With a constant packet rate of 2000/s Intels card could
> capture
> packet headers of short packets but suffered packet loss when capturing
> headers
> from larger packets. The Tulip card showed no losses here. When capturing
> full
> packets, the Intel chip lost up to 35% (1500 Bytes) and the Tulip chip
> showed
> loss of 0.04% or packets larger than 600. This is just to show that
> network card
> & driver has a large influence on your outcome. Also: max granularity of
> timestamps 
> is 10 microseconds, may be more.
> 
> Maybe that helped, 
> Regards,
> Johannes
> 
> On Wednesday, 1. May 2002 14:27, you wrote:
> > Sorry, I just resend this email, to know if somebody has answers to my
> > question.
> >
> > Regards,
> >
> > Syari Mujahid
> >
> > > -----Original Message-----
> > > From:	Mujahid Syari
> > > Sent:	mercredi, 24. avril 2002 16:10
> > > To:	ethereal-users@xxxxxxxxxxxx
> > > Subject:	[Ethereal-users] What is the peak limit for Ethereal ?
> > >
> > > Hi there,
> > >
> > > I am currently testing the Ethereal on the Linux machine to know the
> > > limit of Ethereal (I am using tethereal). At certain level I could
> > > see that Ethereal (or maybe on the kernel level) discards some
> > > packets. When we reached 2.0 Mbyte per second (I use gkrellm on the
> > > source machine), that's when I saw that it discards lots of packets.
> > > I am using Pentium 4 1.7 Ghz with 512 Mb SDRAM 133 Mhz. The network
> > > card is 3COM Fast EtherLink 10/100. Maybe I have done something
> > > wrong (or my number is wrong) somewhere, but
> > >
> > > My questions are :
> > >
> > > *	Has anybody done a benchmark already for Ethereal.
> > > *	If yes what machine are you using (CPU speed, RAM, OS, etc.) and
> > > what is the peak limit for the ethereal before it discards packets.
> > > *	Anybody could tell me what I could do to tweak Linux to make it
> > > runs faster (a part from killing all those unnecessaries processes)
> > > or maybe a site that I could visit.
> > >
> > _______________________________________________
> > Ethereal-users mailing list
> > Ethereal-users@xxxxxxxxxxxx
> > http://www.ethereal.com/mailman/listinfo/ethereal-users
> 
> -- 
> Institute of Communication Networks and Computer Engineering (IND)
> University of Stuttgart
> Pfaffenwaldring 47                           Tel: +49-711-685 7968
> D-70569 Stuttgart                            Fax: +49-711-685 7983
> Johannes Faerber                      faerber@xxxxxxxxxxxxxxxxxxxx