To:
dnsop@cafax.se
From:
"D. J. Bernstein" <djb@cr.yp.to>
Date:
10 Feb 2002 17:01:48 -0000
Automatic-Legal-Notices:
See http://cr.yp.to/mailcopyright.html.
Content-Disposition:
inline
Sender:
owner-dnsop@cafax.se
Subject:
Re: SRV records - when?
ed@alcpress.com writes: > Web browsers are > not smart enough to move on to the next address should the > web server at the first/selected address not be reachable. Really? Which browsers? Have you pointed out RFC 1123, section 2.3? This is not a SRV-specific issue. In my experience, web browsers _do_ connect to multiple addresses, but users give up in annoyance before the second address is tried. This is why tcpclient (http://cr.yp.to/ucspi-tcp/tcpclient.html) uses a fast connection strategy: try all the addresses with 2-second timeouts, then try again with 58-second timeouts. Try asking browser authors to adopt the same feature. The benefit to users is obvious. [ load balancing on the server side ] > The short TTLs that are required cause more bandwidth to be consumed. ``Premature optimization is the root of all evil.'' How many bytes of DNS traffic do you see every day? How many bytes of HTTP traffic do you see every day? ---D. J. Bernstein, Associate Professor, Department of Mathematics, Statistics, and Computer Science, University of Illinois at Chicago