[nsp] slb predictor

[Rubens Kuhl Jr.]

Field experience told me that leastconns can be a nightmare under extremely
high load, because it would take down every node, one by one, by throwing
too much requests at it. If you can impose some boundaries such as max
connections and max new connections/s on a per node basis, than using
predictors such as leastconns or minimum response time will be fine.
Otherwise, stay with something unrelated to the system state such as
roundrobin.


Rubens

----- Original Message ----- 
From: "Daniel Medina" <medina at columbia.edu>
To: <cisco-nsp at puck.nether.net>
Sent: Tuesday, November 11, 2003 8:08 PM
Subject: [nsp] slb predictor


> Are there any important differences between using roundrobin and
> leastconns as slb predictors?
>
>  I was assuming that using leastconns would be preferred in all cases,
> as reals coming into a serverfarm would fill up more quickly.  What's
> the advantage of roundrobin?
>
>  Cisco doesn't seem to offer any documentation about this (beyond
> "leastconns...directs network connections to the server with the fewest
> connections.")
>
> -- 
> Dan
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>