NANOG 40 agenda posted

Mon Jun 4 20:36:28 UTC 2007

On 4-Jun-2007, at 02:03, Colm MacCarthaigh wrote:

> On Mon, Jun 04, 2007 at 02:53:52AM +0000, Paul Vixie wrote:
>
>>> ipv6 load balancers exist, one's current load balancer is/may  
>>> probably
>>> not be up to the task.
>>
>> my favourite load balancer is OSPF ECMP, since there are no extra  
>> boxes,
>> just the routers and switches and hosts i'd have to have anyway.
>>
>> quagga ospf6d works great, and currently lacks only a health check  
>> API.
>
> If you're load-balancing N nodes, and 1 node dies, the distribution  
> hash
> is re-calced and TCP sessions to all N are terminated simultaneously.

Yep. This is a disadvantage that was mentioned in both <http:// 
www.nanog.org/mtg-0505/abley.cluster.html> and <http://www.isc.org/ 
pubs/tn/isc-tn-2004-1.txt>. I seem to think there's general text  
about this in RFC 4786, too. From the ISC tech note:

    CEF's route selection algorithm is stateless and deterministic for a
    stable set of ECMP routes. In general, however, a change in the
    number or ordering of those routes may cause the route selected  
for a
    particular (source, destination) hash to change. This fragility
    should be considered when gauging whether this load distribution
    approach is appropriate to particular protocols.

I have used dedicated load-balancing appliances for this kind of  
application. They have the disadvantages that (a) they are not cheap,  
and (b) sometimes the non-cheapness encourages people to use them in  
a fashion which exposes a single point of failure. They have many  
advantages, too, including (often) a sufficiently-capable state  
engine that the issue you mention does not arise.

As with all things, the trick is to weigh the risk of disaster  
against the probability of benefit and do whatever makes sense within  
your own particular constraints.

Joe