Re: Piranha

Bonjour,

Voilà, je suis de retour, et je continue sur Piranha. J'étais absent.
J'ai une petite question, si c'est possible :

Sébastien Bonnet a écrit :

> > Hi Mustapha,
>
> > We would like to change the wlc scheduling algorithm to take
> > into account, CPU loads, memory usages, and network load of the
> > real servers.
>
> OK.
>
> > The approach is to periodically cheek these parameters, and
> > update the weight assigned to each real server, according to a
> > given function.
>
> So, your problem/question is not about a scheduler but more likely about
> the monitoring tools. LVS by itself does *never* change RS weights. It
> *can* be the monitoring tools' job.
>
> > A first such approach has been made by Jeremy Kerr, that take
> > into account CPU loads, in LVS.
>
> Yes, I know. What I've understood from what he said is that is
> developping a framework that will help anybody easily plug his own
> program do monitor as wanted.
>
> > Yes,  I have to include monitoring tools in the Piranha software.
> > This is why I am looking to the source files in order to have a
> > global comprehensive view of the software.
>
> OK, now it sounds clear :)
>
> > So, could you please be more accurate ?. In particular, about the
> > relation between the different components of the software.
>
> OK, I'll give it a try. When pulse is started, it tries and sees if
> another pulse is running on the other box. If not, it then starts lvs,
> which then starts one nanny for each RS.
>
> By the same time, lvs adds all active virtual servers (not RS) to IPVS.
>
> Nanny polls a real server and add it to (remove it from) IPVS, and
> regularly adjusts its weight based on CPU load.
>
> This is where you should hack (see getRemoteLoad in nanny.c). Usually,
> the load command is uptime, rup or ruptime, but you can specify whatever
> you want as long as the output is similar to what uptime provides.

Au fait, la mise à jour des poids des serveurs réels est faite
indépendamment
de la  technique de scheduling ?. C'est bien ca, au niveau de nanny, donc
de Piranha
et non pas au niveau du noyau IPVS ?

Merci Beaucoup,
Mustapha