[JDEV] Heartbeat patch for dialup and laptop users and faulty presence info

[Dave Turner]

On Wed, May 29, 2002 at 09:55:28AM -0400, Nathan Sharp wrote:
> The biggest argument I've heard so far is that ping/pongs would take too 
> much bandwidth.  If your end users would prefer very slightly less 
> bandwidth used yet LOST MESSAGES AND FAULTY PRESENCE info, well, you got 
> different users than mine.

With the ping/acks there's a simple way to cut down on a large amount of
the bandwidth required.  From your first post I think that you are
already doing this, but I'll just write it down to make sure.

The server only needs to send the ping to the client if it hasn't heard
from the client within the chosen time-period.  Therefore, each time the
server receives a packet from a given client it can reset its timer for
that client.  For active clients this could mean never receiving a ping
because the server is happy that it's alive.

For clients that are inactive and responding to pings the bandwidth
consumed is almost certainly less than that which would be used if the
client has active.  So in the calculations of bandwidth usage I think
you need to include the estimations for how much bandwidth COULD be
used based on the number of connected clients.

Finally, in the implementation the overhead for tracking timeouts for
each client isn't that great.  At first guess one might have a timer
for each client, that's not the way to go.  Use one timer and a
scheduler to queue up future pings, rescheduling when the server
handles a packet for a client.


I'm not sure if pings get my vote yet, but I think this helps show that
the bandwidth argument needn't be too much of a concern.

-- 
Dave Turner
http://figroll.com/