Forum Moderators: phranque
I find myself doing this at moment in order to get maximum performance out of a TCP/IP based system (not actually a webserver, but the technology is largely the same).
Solid article on this here:-
[isi.edu...]
Curious to hear any results from people that have experimented with such low level tweaking. My current objective is to reduce a large volume of TCP/IP socket connections that are sitting in a WAIT_STATE pending an ACK from a client that will never arrive (because the client has gone offline). It seems there are proposed mechanisms for achieving this and having the client handle the WAIT_STATE rather than the server.
I'm not sure if Apache is already doing such things, haven't delved into that code yet.
Anyone here messed around with this?
Is anyone here actively tuning the TCP/IP stack on their webserver OS ?
We'll take that as a "no", shall we? ;) Have you had any good results so far with your tweaking?
Have you had any good results so far with your tweaking?
It's been, let's say, interesting. I've learned a lot about the lower level workings of TCP/IP, SYN and ACK signals etc and how they can get delayed but not ignored (part of the protocol design).
But I can't say that I've found anything particularly useful just yet.
I would use the ideas from Yslow first before trying these.
I'm actually tweaking a TCP/IP based application which isn't for the web. I thought of asking because of course HTTP servers are essentially TCP/IP based. I'll take a look at Yslow though - there might be some useful config tidbits in there - thanks for the heads up.
you could switch from TCP to UDP.
Not an option. UDP breaks NAT and Firewall traversal (if user can access my webpage, I know that TCP/IP port 80 will work).
