From: Eckehardt Luhm <bselu@web.de>
To: linux-kernel@vger.kernel.org
Subject: Network packet drop?
Date: Mon, 19 Nov 2001 08:20:44 +0100 [thread overview]
Message-ID: <3BF8B2CC.C172F67C@web.de> (raw)
Hello.
I'm studying at the University of Karlsruhe and I'm doing some network
measurements. In conjunction with network drivers for Linux, I'm
experiencing very strange effects.
In my setup I have two PCs each with a EEPro-100 network card in it
connected via an X-link cable. mii-tools says they auto-negotiated a
speed of 100baseTx-FD. Furthermore I have a program which generates UDP
packets at the highest possible speed, simply by looping a
sendto(socket, ...), which should block until a packet is out. What I
want to measure is the maximum output the network card can do. So I'm
sending n packets and divide the time needed for that by n.
This works perfectly for larger packets (>500 bytes). All of them are
being sent out. I'm able to verify this by invoking "ifconfig eth0|grep
TX" and watch the transmitted packets grow by the number of packets I
intended to send (in fact it grows a bit more, because of some
ARP-packets, but that doesn't matter).
But when I set the packet size to say 50 bytes (only data size without
any headers, so on ethernet this would be 102 bytes), something strange
is going on. Now not all of the packets I send to the other network card
are being received. There are leaks of tens of packets at the receiver,
I'm verifying this with tcpdump. Ok, the receiver was just overtaxed, in
the syslog I got "eth0: Abnormal interrupt, status 00000002.". So it
couldn't handle the flood of packets, ok.
BUT: The sender didn't even send all of the packets! "ifconfig eth0|grep
TX" grew only about 40-50% of the value it should! e.g. I sent 10.000
packets, and the TX-counter grew only by 4586, not more. Sometimes I got
even worse results, especially when decreasing the packet size.
I tried several setups on different PCs, sometimes with X-linked network
cards, sometimes with a switch between them. None of them worked. Except
one thing: I tested the same setup described above with other network
card, two noname products with a realtek 8139 chipset (driver module
8139too). And you would have guessed it: That worked! The TX-counter
grew by the correct value, so all packets have been sent out.
And to make the confusion perfect: With forcing the cards to work with
10 MBit/s-FD (with mii-tool), the same strange "packet drops" as with
the EEPro-cards appeared.
Is there anybody out there, who can explain what is going on in network
drivers? What causes these strange effects?
Regards, Elu
--- Eckehardt Luhm, eMail: bselu@web.de
next reply other threads:[~2001-11-19 7:51 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2001-11-19 7:20 Eckehardt Luhm [this message]
2001-11-19 8:32 ` Network packet drop? Zhongyu
2001-11-19 10:00 ` Alan Cox
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3BF8B2CC.C172F67C@web.de \
--to=bselu@web.de \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox