Linux Advanced Routing and Traffic Control list
 help / color / mirror / Atom feed
From: Grant Taylor <gtaylor@riverviewtech.net>
To: lartc@vger.kernel.org
Subject: Re: [LARTC] multipath device round robin not working?
Date: Mon, 15 Jan 2007 04:14:32 +0000	[thread overview]
Message-ID: <45AAFFA8.1010709@riverviewtech.net> (raw)
In-Reply-To: <45A8C870.2010806@gmail.com>

On 01/13/07 05:54, zutph3n@gmail.com wrote:
> Both links have their own ip but have the same gateway. The problem is I 
> can't seem to get egress traffic load balanced over the 2 nics.

I don't know if it is still a problem or not, but I ran in to something 
very similar a LONG time ago (mid 2.4).

Basically what I found was the problem was that (I believe) multi-path 
routing really is multi gateway routing.  I.e. load balancing across two 
(or more) different gateways.  In your case, and the case that I had, 
all the IPs in the world did not make any difference b/c each path had 
the same default gateway.

My solution at the time was to use UML routers to provide different 
subnets to the box doing ECMP routing.  Each UML basically NATed from 
the one upstream network to a small downstream private subnet that was 
unique for each link.  This allowed the box doing ECMP to see different 
gateways.  This worked great until I hit a memory limit on connection state.

I should probably say that the problem I ran in to was not a problem 
with ECMP but the number of hosts that I was trying to NAT with the 
amount of RAM that was on the box.  I was able to resolve this by adding 
RAM to the box to provide a larger Connection State Table.  For the 
record, the box was running a mid 2.4 kernel with a subnet that was the 
size of 4 class C networks (2048 IPs) on a box that had 256 MB of RAM. 
I ended up taking the box up to 2 GB of RAM and things have been working 
GREAT ever sense.  I do believe this memory / connection state problem 
has been resolved long ago.  However the system is working and payed for 
and the client is perfectly happy with what is in place and sees no 
reason to do any thing with it.  (If any one would like more details, 
just ask.)



Grant. . . .
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

  parent reply	other threads:[~2007-01-15  4:14 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-01-13 11:54 [LARTC] multipath device round robin not working? zutph3n
2007-01-14  9:26 ` Alex Samad
2007-01-15  4:14 ` Grant Taylor [this message]
2007-01-16  0:44 ` Grant Taylor
2007-01-16 19:52 ` Alex Samad
2007-01-17  5:04 ` Grant Taylor

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=45AAFFA8.1010709@riverviewtech.net \
    --to=gtaylor@riverviewtech.net \
    --cc=lartc@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