httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexei Kosut <>
Subject Re: Keepalive netscape bug recap
Date Mon, 15 Apr 1996 23:09:31 GMT
On Mon, 15 Apr 1996, Rob Hartill wrote:

> Can someone (Alexei?) remind me of the problem with Netscape. I rememeber
> something about a performance hit caused by flushing output to keep
> a broken Netscape happy.

That's about right. There were two problems:

Netscape Navigator 2.0 (and possibly later versions) have a bug
where if the end of the body of a response message was in the same
(256-byte) read as a header, then it would not properly return the
document to user control - it would hang until the kept-alive
connection timed out.

We fixed this by adding an fflush() after the headers were sent,
causing Netscape to read the body on another read, but this caused a
performance decrease, as an fflush() takes a while on most systems. We
ended up only flushing for persistent connections, which is okay,
assuming that the benefit from the persistent connection is greater
than the slowdown from the flushed buffer (which it probably is).

Then there's Ben's Windows bug... but that's something else.

Alexei Kosut <>
Lefler on IRC, DALnet <>

View raw message