Re: Routing 600+ vlan's via linux problems (looks like arp problems)

[Willy Tarreau <w@1wt.eu>]

On Thu, May 03, 2007 at 10:25:48PM +0200, Øyvind Vågen Jægtnes wrote:
> Hi,
> 
> We have a one gigabit internet connection that is normally
> routed by a hardware juniper router. The drive in this is down
> and we need to use a linux machine (Pentium D 3 ghz) as a
> temporary router.
> Now setting up all the 600 vlans and assigning ip addresses
> is no problem. We have testet all by using a laptop, setting up
> 600 vlan interfaces on this and running dhcpclient on all.
> This worked just fine, all the interfaces got address.
> 
> Now for the real setup.
> We closed the mac of the juniper to the network card that
> would be connected to the internal LAN, set up the interfaces,
> and swapped cables. This worked fine for approximately 100
> of the computers that are connected, but the rest would not
> get IP. The connected 100 computers were routed just fine.
> 
> What we think the problem is, is that the arp cache on the
> linux router seems strange. It can resolve the MAC for the
> 100 clients that actually got through.
> For the rest all we see in the arp cache is (incomplete)

I suspect that your arp cache is full (128 entries by default).
Check /proc/sys/net/ipv4/neigh/gc_thresh1 (128 for me). You can
set it as high as gc_thresh2 (512 for me), and I don't know what
happens above.

Hoping this helps,
Willy