[6947] in North American Network Operators' Group
Re: Questions about Internet Packet Losses
daemon@ATHENA.MIT.EDU (Vern Paxson)
Tue Jan 14 04:56:40 1997
To: Tony Li <tli@jnx.com>
Cc: amb@xara.net, nanog@merit.edu
In-reply-to: Your message of Tue, 14 Jan 1997 01:42:35 PST.
Date: Tue, 14 Jan 1997 01:55:19 PST
From: Vern Paxson <vern@ee.lbl.gov>
> Isn't it unsurpsrising that >40% of packets are small? ACKs?
>
> If one made the gross oversimplifying approximation that everything is
> unidirectional TCP traffic, then you'd expect to see one ACK per two data
> packets.
This is actually not at all a gross oversimplying approximation. Much
TCP traffic unidirectional or request/response. The latter looks a whole
lot like two back-to-back unidirectional connections, with only one chance
to piggyback acks at the switchover.
> Thus, we'd expect to see 33% at 40 bytes. It is the additional
> 7+% that's surprising.
Throw in frequent SYN/FIN/RST's because of small Web connections, SYN
retransmissions because busy Web servers have full listen queues (or are
being SYN-flooded :-), TCP's that ack every packet (Linux 1.0, I believe,
and maybe later ones), TCP's that wind up acking every packet because their
delay timer is less than an MSS prop time across a slow link (Solaris, for
links < 10KB/sec), and dup acks due to sequence holes, and you might
handwave the 7%.
Vern
