- From: Koen Holtman <koen@win.tue.nl>
- Date: Mon, 12 Aug 1996 23:08:30 +0200 (MET DST)
- To: Paul Leach <paulle@microsoft.com>
- Cc: koen@win.tue.nl, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com, hallam@etna.ai.mit.edu
Paul Leach: >>From: koen@win.tue.nl[SMTP:koen@win.tue.nl] >>>2. It was taken between proxy and server, >> >>This is because the internet backbone links are the bottleneck, not >>the LAN between your proxy and your user agent. Who cares if you get, >>say, 30% savings in web traffic on the LAN? In this game, it is >>backbone savings that count. > >Not everyone has a LAN link between user agent and proxy. Some small >number of people dial in to their proxy, which is owned by their >service provider. Say order 10 million or so, with providers like AOL >and MSN. I'm not a specialist in modem technology, but AFAIK modems offer you a constant amount of bandwidth both ways. So savings on the request stream will not get you more bandwidth on the response stream, which means that the savings matter even _less_ than on a LAN. My wait chain calculations will _over_estimate the speedup in the modem case, assuming that the modem is the bottleneck. As for the high RTT on some of Digital's internal links: RTTs are irrelevant for this discussion. If anything, they will make the savings less noticable. If Digital's internal links were highly _saturated_, that would be another thing. [...] >>And the modem hooked to the ordinary telephone line will probably do >>data compression, so you would gain little extra with sticky headers, >>I believe. > >In your model, a 200 byte request would be reduced by useing sticky >headers to 40 bytes, and then the 40 bytes would be compressed to (say) >20 bytes (assuming a 2-1 compression ratio). Without sticky headers, and >the same compression ratio, requests would be 100 bytes. The savings >from sticky headers is still the same 80%. The compression ratio varies depending on what you try to compress. Have you ever gzipped a httpd logfile? If the same effect happens with modem compression, Jim will learn something about how HTTP works very soon. >Paul Koen.
Received on Monday, 12 August 1996 14:12:22 UTC