[aprssig] Periodic Disconnects from APRS-IS
gregg at wonderly.org
Fri Jan 19 09:31:27 CST 2007
AE5PL Lists wrote:
> This is not an issue of computer buffering but of TCP buffering.
Yes Pete, but when there is a flood of data, it make sense to send as much data
as you have in a single packet. Those return ACKs are part of the overall
latency applicable to the delivery of data between machines. So, the fewer acks
there are, the more opportunity there is for increased bandwidth utilization.
Also, it occurs to me that most consumer highspeed internet services are
asymetrical in bandwidth, and those ACKs go out through the slow side.
This specific issue is what my sample algorithm was about. Write as much as you
have, without any delays. Setting the socket TCP_NODELAY, is an additional step
that is independent of this output optimization. The larger packets will not
have a negative impact on TCP_NODELAY. It will simply reduce the number of ACKS
needed and since they go through the lower bandwidth side of the interface, that
might be the turning point for some users between consuming a full feed and not.
More information about the aprssig