> >>>>> "Roger" == Roger Marquis <[EMAIL PROTECTED]> writes: > > Roger> Actually, it seems kind of silly to use TCP for what is essentially a > Roger> connectionless protocol. We do need reliability, of course, but that > Roger> can be taken care of with application-generated ACKs, without a lot of > Roger> coding. We don't need the sequence numbers, sliding windows, or other > Roger> traffic shaping features of TCP. We also don't need to initiate > Roger> connections from the destination to the source. > > That last point is _highly_ debateable. I have a strong desire for a pull > mechanism, as well as a push mechanism. Push or pull, the transport protocol would/might be the same. Implementing this might be a task for later? Rethinking this point. I agree. If a central logger system is developed, it may be optimal to have a pull system. The remote systems might need to understand that they will serve on-demand instead of on-message. The queuing mechanism would have to be looked at. Especially if the wake up and feedme call came down a serial line?? The tcp_wrapper/inetd combo on in.syslog2d would be very good -- this does not cover serial lines... Rob
