Unfortunately the list archives are not searchable, so
maybe this was already discussed:

In the Slashdot article "The trouble with the Nagle
algorithm"
(http://developers.slashdot.org/comments.pl?sid=174457&threshold=1&commentsort=0&mode=thread&cid=14515105)
by John Nagle himself, it is recommended against
performance losses: "The user-level solution is to
avoid write-write-read sequences on sockets."   

Unfortunately libneon does exactly this by writing out
first the http header and then the data:

ne_request.c:

send_request()
        -> ne_sock_fullwrite()
        -> send_request_body()

Of course, you can use TCP_NODELAY. But wouldn't it be
better in the first place to not write the header and
data separately?

When first using neon, I spent some time wondering why
a neon http call took 300 ms until I figured out to
set TCP_NODELAY (which got the time down to 10 ms).

Juergen
 

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
neon mailing list
[email protected]
http://mailman.webdav.org/mailman/listinfo/neon

Reply via email to