On Mon, Jun 05, 2000 at 10:37:42AM +0200, Rogier Wolff wrote:
>
> Isn't it kind of odd, that in the 20 seconds between :20 and :40 the
> original packet and something like 3 retransmits were lost? Would
> TCP/IP allow bwww to reset the retransmit timer one notch down if it
> receives the ack 924 ?
This is possible with SACKs: the receiver tells the sender that there is
a hole in the sequence space and the sender retransmits it quickly.
You don't see SACK use here because 2.0 didn't support it.
Of course again it only works when there are enough packets in flight,
it may not help for telnet packet at all (maybe you should look at implementing
line mode in ssh / telnetd instead ;)
>
> (TCP/IP algorithms are based on the "fact" that packetloss is caused
> by overloaded routers. In my case that often isn't the case. I now
> found a cable that I could use that generates 20% packetloss, but my
> cable-company seems to find 5-10% packetloss in their network
> acceptable, and it isn't related to congestion or anything.
> Retransmitting every 3 seconds (no exponential backoff) would be the
> best strategy to get "things done" under these circumstances).
ECN may fix this partly, but it'll take a long time until all routers support it
(Linux 2.4 does at least)
-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:20 EST