Here's a hypothesis: According to the paper, a typical connection is at a
(surprisingly) low rate: 14-18 packets over 10-15 seconds. This means that
packets are arriving at the client at roughly a rate of about 1-2 packets
per second. Say the client gets a packet at time t. Then the delayed ACK
timer (p. 265-267 in Stevens vol. I) will probably expire at or before
time t+200ms, before the next packet arrives at time t+500ms to t+1000ms.
At this time, the client sends an ACK. Thus, instead of waiting to ACK
every other packet, every packet is ACKed, This seems like the right thing
to do if the data really is trickling in like that.
This is consistent with their results where almost half of packets on the
web are 40 bytes - presumably mostly these symmetric client ACKS plus a
few SYNs, FINs.
Of course with these numbers, we're talking 1KB/s for a web object
download. This seems surprisingly low, though perhaps not if we believe
the factoid that 70% of web clients are over modems. Also, i don't think i
understand the ack-every-other-data-segment mechanism of TCP; is this
separate from the delayed ACK feature?
neal
X-VM-v5-Data: ([nil nil nil nil t nil nil nil nil]
[nil "Wed" "29" "April" "1998" "17:36:35" "-0700" "Neal Cardwell" "cardwell@cs.washington.edu" "<Pine.LNX.3.96.980429173440.28071N-100000@sake.cs.washington.edu>" "35" "Re: Fwd: Re: progress " "^From:" nil nil "4" nil nil nil nil]
nil)
Received: from sake.cs.washington.edu (sake.cs.washington.edu [128.95.4.55]) by june.cs.washington.edu (8.8.7+CS/7.2ju) with ESMTP id RAA17423 for <voelker@cs.washington.edu>; Wed, 29 Apr 1998 17:36:36 -0700
Received: from localhost (cardwell@localhost) by sake.cs.washington.edu (8.8.8+CS/7.2ws+) with SMTP id RAA29737 for <voelker@cs.washington.edu>; Wed, 29 Apr 1998 17:36:35 -0700
In-Reply-To: <199804292011.NAA40316@joker.cs.washington.edu>
Message-ID: <Pine.LNX.3.96.980429173440.28071N-100000@sake.cs.washington.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
From: Neal Cardwell <cardwell@cs.washington.edu>
To: Geoff Voelker <voelker@cs.washington.edu>
Subject: Re: Fwd: Re: progress
Date: Wed, 29 Apr 1998 17:36:35 -0700 (PDT)
Looks like I can't log into sirpa-a yet. Who should i talk to about that?
David Becker?
thanks,
neal
On Wed, 29 Apr 1998, Geoff Voelker wrote:
> ok, until we get into the amd maps as /projects/networking, we'll have
> to work on sirpa-a. I chgrp'd /loom/syn and /loom/syn/www into the
> syn group (it only shows as a number because it's not in sirpa-a's
> group list, but it will show correctly when mounted on normal machines
> -- what a terrible name for a machine, btw) and made it group
> writeable. so you should be able to put files there and have them
> seen by the web server.
>
> the file system isn't being backed up yet, though; I'll ask for that
> once the /projects map is updated (have to wait for that anyways).
>
> -geoff
>
> ------- start of forwarded message (RFC 934 encapsulation) -------
> From: Scott Rose <rose@cs.washington.edu>
> To: Geoff Voelker <voelker@cs.washington.edu>
> cc: cardwell@bauhaus.cs.washington.edu
> Subject: Re: progress
> Date: Wed, 29 Apr 1998 13:04:04 -0700
>
>
> My mistake. It's serving fine now.
>
> > I created an index.html, but I got a 404 when trying to load under netscape.
> ------- end -------
>