[Spread-users] Re: Hi to everyone and a question

[Jens]

I'm replying to my own questions :)

I have done a bit more homework and found some answers, in the mailing
archives.
The token protocol was mentioned discussing Mike Perik's performance issues.
If I understand it correctly, in an idle situation the token is held by the
"leader".
The leader seems to be the first machine listed in the spread.conf file.
When the token
is needed by another daemon it's requested from the leader. When that daemon
is
done sending the token will circle for a while until a configured number of
rounds
without anyone sending data. Then it will fall back to the leader.

What wasn't clear is if there is one token per network or one token per
"domain" of
spread daemons or if there is one token per group. I would assume the worst
that there
is only one token per domain of daemons.

Same as Mike, I'm evaluating if spread is a useful middleware for financial
data
distribution for market data. In such an application one would have one or
more
heavy senders and many rather passive receivers. The spread daemons would be
distributed over at least 2 main networks connected via a leased line (main
and
backup site). Low latency of the market data transmission is critical thus
the token
 concept has some drawbacks since senders have to wait before they can send
data.
Another issue is that a single "faulty" machine/network node can drag down
the
performance of the entire set of spread daemons. Faulty could mean a broken
network
card, cable, switch port, or just a (CPU) starved spread daemon. As I
understand it,
that one machine can "hog" the token for extended periods and hence prevent
other
nodes from sending urgent data.

Maybe someone could confirm/correct my understanding of how the token
algorithm
works.

Thanks

Jens