[j-nsp] Multihoming servers to two Virtual Chassises

[Tore Anderson]

Hi,

I'm about to deploy a stack of EX 4200 switches in a new data centre.  
My original plan was to make a Virtual Chassis, split it in two stacks 
and put those stacks in separate corners of the room, and then let the 
servers (running almost exclusively Linux) multi-home to both stacks 
(using 802.3ad).  In addition to this the switches will function as 
distribution switches for more peripheral parts of the network.

After playing around with the switches in a lab I'm not so sure if I 
want this after all.  If the switch functioning as routing engine 
fails, the switchover to the backup takes too long, and the OSPF 
sessions to the upstream MX 240-es are also lost.  This happens with 
GRES active and BFD off, too.  So a RE failover causes a 30-40 second 
L3 outage.  L2 forwarding seems unaffected, though.

This isn't acceptable to me, so right now I'm looking at a setup where 
the stacks in each corner are separate VCes, which should ensure 
minimal L3 downtime if one fails - with BFD protection on both the up- 
and downstream OSPF sessions all traffic should be rerouted over the 
remaining VC within seconds.  Two active REs will also give me some 
additional resiliency in case of JUNOS bugs, so I quite like this 
solution.

However now I have a problem with the multihomed servers.  Is it 
possible to have a 802.3ad LAG span multiple VCes (à la Nortel SMLT)?  
If not, does the EXes have any other nifty feature that could help me 
accomplish multihoming servers like that?  The doubling of bandwidth is 
not essential - high availability is what I need to accomplish, so 
simple failover is fine.

I appreciate any suggestions you might have.

BTW:  Anyone have an idea of when (if?) the EXes will support IPv6?

Regards,
-- 
Tore Anderson