[1221] in linux-net channel archive
Re: bug? Linux drops some broadcasts...
daemon@ATHENA.MIT.EDU (Vincent Cojot)
Tue Oct 17 20:29:47 1995
Date: Tue, 17 Oct 1995 14:35:29 -0400 (EDT)
From: Vincent Cojot <coyote@step.polymtl.ca>
To: linux-net@vger.rutgers.edu
Hi everyone,
First, I'd like to thank publicly all those who replied to my
desperate message. I will try to answer these mails in order. (btw: can
anyone tell me how to subscribe to this mailing list and most importantly
if, being only a user and not a developper, I am allowed to subscribe to
it). Second, i'd like to apologize for the length of this message.
Jeff Noxon <jeff@router.patch.net> writes:
> Try using gated on Linux. The linux routed port (if you can call it that)
> is incredibly buggy. If you need diffs, I will send them to you.
>
> Jeff
They are reluctant to run gated for a couple reasons, one of them
being that since the school's main inner gateway runs gated as well, if
that machine goes down, the lab's gateway (the linux machine) would try to
compensate and try to route everything by itself (which it can't probably
handle). [I'd like to have your diffs, too, if possible.. :) ]
***********************************************
Arnt Gulbrandsen <agulbra@troll.no> writes:
> You need to specify:
>
> 1. Your hardware configuration.
Intel enadeavour MB, 3 SMC ULTRA cards (the kind with a real eeprom, not
those ether-ez), 2 2940W adapters. 64mb ram, 4 2GB disks, the computer is
clean IRQ wise (I helped check that myself so I guess it's not a hardware
conflict, unless it's something non standard happening).
ifconfig -a reports (see maps of the local net below in the next message):
lo Link encap:Local Loopback
inet addr:127.0.0.1 Bcast:127.255.255.255 Mask:255.0.0.0
UP BROADCAST LOOPBACK RUNNING MTU:2000 Metric:1
RX packets:3446 errors:0 dropped:0 overruns:0
TX packets:3446 errors:0 dropped:0 overruns:0
eth0 Link encap:10Mbps Ethernet HWaddr 00:00:C0:61:58:8F
inet addr:132.207.4.4 Bcast:132.207.4.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:119790 errors:0 dropped:0 overruns:0
TX packets:117928 errors:0 dropped:0 overruns:0
Interrupt:10 Base address:0x2b0 Memory:dc000-e0000
eth1 Link encap:10Mbps Ethernet HWaddr 00:00:C0:7F:5A:8F
inet addr:132.207.12.2 Bcast:132.207.12.127 Mask:255.255.255.128
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:139224 errors:0 dropped:0 overruns:0
TX packets:162065 errors:0 dropped:0 overruns:0
Interrupt:5 Base address:0x290 Memory:d8000-dc000
eth2 Link encap:10Mbps Ethernet HWaddr 00:00:C0:86:F8:7F
inet addr:132.207.12.129 Bcast:132.207.12.255 Mask:255.255.255.128
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:121672 errors:0 dropped:0 overruns:0
TX packets:91647 errors:0 dropped:0 overruns:0
Interrupt:3 Base address:0x250 Memory:d4000-d8000
> 2. Whether tcpdump on the linux box sees all the packets.
Yes, tcpdump on the linux machine sees all four broadcast packets.
> 3. Whether tcpdump on another box on the same ethernet segment sees
> all the packets.
Yes, this was checked.
> --Arnt
**************************************************
Jim Shankland <jas@flyingfox.COM> writes:
> Maybe; however, rather than Linux in general, I'd take a close
> look at the network card you're using. Some cards work better
> than others, and some have better Linux drivers than others.
> For what it's worth, I've had pretty good luck with the
> 3C509 cards.
I will try to get some of these for test purposes (it's what we have in
the linux workstations but the server has 3 smc cards..).
> But why do you need all these routing updates anyway? If all
> packets pass through the AIX machine, as your picture implies,
> then why not:
Let me draw a more detailed map of the network (I did not do a very
detailed map because I did not want to put too many details in..), here
is what our school looks like (it has about 95 subnets and hundreds of
unix workstations on the net). I am sorry if I made some confusion about
which gateway was which.
******* Lab Lab Lab Lab
LinuxLab Lab Lab Lab Lab Lab | | | |
| | | | | | ------------------------
--------------intrntgw--------GTW--| token-ring backbone |
| ------------------------
|
Internet
We can't put machine "GTW" as the default route for "linuxlab" because it
could overload it and the school's computing services are trying to avoid
that. It's why "GTW" is broadcasting packets saying to which subnets it
is connected and that packets going to these subnets are going through
itself but not others..
putting static routes is not practical because it means they'd have to
manage by hand about 40-50 routes to various subnets inside the school.
> Static routes are not always appropriate, but in this case they are
> (in my opinion) -- unless there are other network interconnections
> that you did not show from the SUNS or PCS net.
Hmmm. In fact the "Linuxlab" looks like:
--------<-----Suns-------<------
| |
VonN (sparcstation 10) ADA--------->-----
| (running routed too) | (pc linux server + gateway for the)
| | (whole lab: suns + PCs)
-------->----Linux-PCs-->-------
What it means is:
- To reach the outside, "Suns" have to go through "VonN" and then through
"ADA".
- To reach the "Suns", the "PCs" have to go through "ADA".
- To reach the "PCs" from the outside, packets have to go through "VonN"
and then through "ADA".
The problem of "dropped broadcast packets in happening on 'ADA'".
**********************************************************
Thanks if you read until there.. I hope we can get this problem resolved,
I am sure a lot of people would be greateful for ever to those who could
help us. As time goes by, some people in the dept.'s admin seem more and
more convinced we should install Solaris but I am not convinced it will
solve their problems.. :) but then it would be really nice if it could
work fine with linux..
Vincent S. Cojot
coyote@step.polymtl.ca
