Re: Release new code: Scheduler for distributed caching

To: Joe Cooper <joe@xxxxxxxxxxxxx>
Subject: Re: Release new code: Scheduler for distributed caching
Cc: "Matthew S. Crocker" <matthew@xxxxxxxxxxx>, lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Lars Marowsky-Bree <lmb@xxxxxxx>
Date: Fri, 27 Oct 2000 11:46:05 +0200
On 2000-10-27T04:41:32,
   Joe Cooper <joe@xxxxxxxxxxxxx> said:

> An API for detecting load within the real server and reporting it to the
> director is a good idea.  The only problem is that it isn't quite as
> simple a solution as it sounds.

Well, the API is quite easy.

Report current load (percentage, however you want to calculate that) and
currently active connections.

You may think "currently active connections" is something the director knows
anyway. Not so, if you are using multiple directors.

>From this, the LVS could for example come up with Current * (1/Percentage-1)
to find the theoretical "free" connections until the server load is 100% or
something, and select the least loaded server.

The API for this - which I would propose - is the heartbeat API. The LVS
clients and servers can register as a service and use it to communicate
key/value pairs, and will also be notified when a node goes up or down.

Initial common fields would be hostname (that is part of heartbeat's
information anyway), load_percentage and cur_connections.

    Lars Marowsky-Brée <lmb@xxxxxxx>
    Development HA

Perfection is our goal, excellence will be tolerated. -- J. Yahl

<Prev in Thread] Current Thread [Next in Thread>