httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From dean gaudet <dgaudet-list-new-ht...@arctic.org>
Subject Re: What is the purpose of saferead?
Date Sat, 17 Jun 2000 19:54:36 GMT
On Wed, 14 Jun 2000, Bill Stoddard wrote:

> It seems to be an optimization for handling pipelined requests. Rather
> than immediately flushing the output buffers after handling a request,
> we do a non-blocking read to see if another request (presumably a
> pipelined request) is on the wire. If another request is read, we
> handle it and the response can be pipelined back to the client along
> with the earlier response. Is this about right?

yup.

the saferead/halfduplex stuff was all added when i was fixing apache to
generate optimal network packets ... for this paper
<http://www.w3.org/Protocols/HTTP/Performance/Pipeline.html>

> Doing the non-blocking read is really causing me a headache with
> Windows. I don't know how to do a non-blocking read on Windows w/o
> using select (which is a performance hog). I can do an async read, but
> managing the i/o buffer on an async read is more than I care to deal
> with right now. Would it be acceptable to bypass the saferead on
> Windows and do an unconditional flush on the output buffer when we are
> done handling the request?

you would generate more network traffic than necessary, so it's up to you.

-dean


Mime
View raw message