Unsuccessful gateway. Too may arp coversations.

Hi,

I am having problems with my new RHAS3 Linux which I have setup as the
gateway for our LAN to internet.

When setting the system up I have used 5-10 local workstations to see
and configure the routing and NAT. Everything was working perfectly. I
also installed squid and saw the benefit of it too.


But now when I configure more 50-100 or above clients to use this
gateway, it works only during the first 10-30 minutes. then the system
stops acting as a gateway.

Both the public and private NIC's do respond to ping, but does not
forward traffic.

I have installed Ethereal on a local workstation and found out that
within 2 minutes, over 300,000 conversations are recorded, and of
99.35% of these are ARP broadcasts. There are over 15 machines each
owning 10,000-35,000 of these conversations.


Can this be the cause ? If yes, how can I fix it ?

Regards.

Check for a loop in the network.

What kind of a loop ?

We have a backbone switch and over 50 side switches connected through
fiber to the backbone, with lugs(stacks) of 4 of them max.

In our lan, there is only one segment (subnet) and a single internet
connection.

What kind of a loop can be ?


On 29 Mar 2005 08:50:12 -0800, "pizzy" <(E-Mail Removed)> wrote:

>Check for a loop in the network.

Sanal Kisi wrote:

[please don't top post -- makes interleaving answers difficult]

> On 29 Mar 2005 08:50:12 -0800, "pizzy" <(E-Mail Removed)>
wrote:
>
> >Check for a loop in the network.

> What kind of a loop ?
>
> We have a backbone switch and over 50 side switches connected through
> fiber to the backbone, with lugs(stacks) of 4 of them max.

Problem may lie here with a misplaced patch cord.

> In our lan, there is only one segment (subnet) and a single internet
> connection.

In a switched network "segment" refers to the parts of the network
separated (ie., segmented) by the switches, so you have many segments.
Think of the switches as multi-port bridges.

> What kind of a loop can be ?

A loop is when you have multiple physical pathways between segments.
This is done intentionally to provide redundancy/failover to critical
segments.

It can happen quite easily even when you don't intend it if you cascade
or string together more than 2 or 3 switches, especially with a
misplaced patch cord. Misconfigured switches can make it worse.

Broadcast frames then get caught in a loop, appearing first on one side
of a segment, then the other, then back to the first, etc. This can
easily lead to broadcast storms.

Sometimes the "looping" behavior shows up when a number of nodes have
been moved from one segment to another without flushing the switches'
MAC tables. Suddenly MAC addresses start showing up on different sides
of the switch.

>From the volume you speak of, it sounds like a physical loop. They can
be hard to locate. Narrow down the probable source or area by
examining the captured frames. You may have to bring different
segments "on line" in a controlled fashion to find which segments'
nodes are generating the initial arps that get caught in the loop.

I'm familiar with Cisco docs, so here are some you can look at:
http://www.cisco.com/univercd/cc/td/...2.htm#xtocid20
http://www.cisco.com/univercd/cc/td/...3.htm#xtocid88

hth,
prg