|
Wensong Zhang wrote:
> * Load-informed scheduling
> The agents running on real servers collect the load and memory information
> and send them to the agent(poller) running on the LinuxDirector, and the
> poller give load information to the ippfvs in the kernel, the ippfvs
> balances packets over real servers according to load information.
>
> +1: matthew, wensong
> -0:
> Status: matthew has implemented this scheduling in the Zaphod project
Hi Everybody,
Does anyone has more info on "Load-informed scheduling" ?
Whay kind of data will be provided back (load data ? new weight ?) and
how (via eth ? serial ? snmp ? etc...)
It would be interesting to know a bit more about it so people could
start writing agents to provide this info back to the load balancer.
Also, I am looking forward adding a "cluster bypass" feature that would
be tied to this "Load-informed scheduling" feature : when the cluster
nodes are all too busy traffic could be redirected elsewhere.
Thanks.
Michel Blanc.
|
