1999-10-20-12:07:45 Roger Marquis:
 > [...] We don't need the sequence numbers, sliding windows, or other
 > traffic shaping features of TCP.

I take it that your vision of syslog is strictly confined to low logging
traffic volumes. There may well be a use for a whizzo clever logger that
crumbles when you fling bigger volumes of data at it, but it's not what I
want. Reliably handling heavy traffic is precisely what the sliding windows
and the traffic shaping features of TCP are there for. I want them in the next
generation logging protocol.

 > We also don't need to initiate connections from the destination to the
 > source.

That strikes me as unrelated to the question of UDP -vs- TCP.

 > The KIS principle should drive this decision.  Which makes a simpler
 > protocol?  Which makes a simpler implementation?  Which will speed
 > syslog2's adoption?  Let's learn from the lessons of SNMPv2 and not
 > repeat them.

Certainly, we want the simplest design that will meet our goals.

Do our goals include being able to reliably and safely log large volumes of
data? If they do, then we want TCP.

-Bennett

Reply via email to