[Beowulf] split traffic to two interfaces for two "subnets"

[Yaroslav Halchenko]

Great great great news for me ;-)

Thank you Joshua -- I will try to set things up and use trunking


On Thu, 11 May 2006, Joshua Baker-LePain wrote:

> On Thu, 11 May 2006 at 11:03am, Yaroslav Halchenko wrote

> >Hi Sean,

> >On Thu, 11 May 2006, Sean Dilda wrote:
> >>>10.0.0.16/28 goes through eth1
> >>>the rest 10.0.0.0/24 can go through eth0 (as before)
> >>Instead of trying to put certain nodes on eth0 and certain nodes on eth1, have you considered bonding eth0 and
> >>eth1 together and letting traffic be spread across them like that?  It should automatically balance traffic for
> >>you.
> >Please please please correct me if I am wrong but I thought that bonding
> >is done between targets, ie all the nodes has to have bonded interfaces
> >to take advantage of bonding?

> Nope.

> >or I can just bond 2 interfaces on the server and leave the rest of the
> >nodes connected with 1 interface to the switch?

> Yep.  I'm doing this with my cluster -- the fileservers have a 2-port bonded interface to the switch, and all the 
> nodes have just 1 connection. I did some quick testing and found that, for me, mode 0 bonding (balance-rr) worked 
> better than mode 4 (802.3ad).

> >I have trunking (which is I believe is the same as bonding) option
> >in my DGS-1248T... may be I should RTFM for the beast...

> Trunking is the same as mode 0 bonding.  I just defined 3 trunk groups (one for each of my fileservers), with each 
> trunk group having 2 member ports.
-- 
                                  .-.
=------------------------------   /v\  ----------------------------=
Keep in touch                    // \\     (yoh@|www.)onerussian.com
Yaroslav Halchenko              /(   )\               ICQ#: 60653192
                   Linux User    ^^-^^    [175555]


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20060511/490e7c9f/attachment.sig>