httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
Subject Re: ap_r patch performance comparison
Date Fri, 19 Jan 2001 06:17:18 GMT

> You inserted a call to ap_rflush() in ap_finalize_request_protocol(). That
> actually flushes the output to the wire (which would obviate pipelining if
> we had a 1.1 connection). So your trace shows the writev() caused by the
> flush, then the server "peeking" for another request.

You're correct.  The ap_rflush was an afterthought, and I believe it can
and should be done another way, I just wan't interested in working too
hard at it.

> You are correct, though: the read should not have happened since we were
> going to close the connection. The check_pipeline_flush() function in
> http_request.c should check the "connection close" flag before doing the
> peek.
> Ah. Check r->connection->keepalive.

Yep.  I never checked any of that, so we will need to add that if
statement to check_pipeline_flush.

> Or you can just commit mine ;-)

You're not going to like my next e-mail (advance warning :-)  I have begun
to review your patch, and I have some questions.


Ryan Bloom               
406 29th St.
San Francisco, CA 94131

View raw message