ip heartbeats

[Tom Vogt]

back again. :)


I've checked a couple more sources, including the heartbeat code from
the website. I've also started a design of my own. please tell me what
you think and especially if you know of projects or even code along
similiar lines.


what I want is basically NOT a complete cluster package, but a low-level
heartbeat function. this might even be part of the kernel (say, a
heartbeat module) the same way that ping is. it's quite similiar to
ping, except that you don't ping someone else to find out whether he's
still there, but broadcast a "ping" to tell everyone else that YOU are
still there.


after checking, I believe broadcasts are the way to go. this is where I
need input, however. several dozen machines continually broadcasting on
the same lan might get problematic - does anyone know what to look out
for?

I do not want to get application stuff into this. which is why I believe
a single udp packet is sufficient, containing not much more than a group
id, the machine's load average and either a time-stamp or a sequence
number (I prefer the later) as data. in addition, a few special tags
might be used e.g. for a system to broadcast when it comes online
(again) first or when it goes down the regular way (shutdown).

finally, a mechanism to take over either ip's or mac addresses. does
anyone have real-life experience which one works better?

and that's it. I'd love to leave everything else to the application
layer, to make this as generic as possible. 


-- 
Tom Vogt

System Manager WLW Online