From owner-freebsd-questions Sat Aug 7 12:28:30 1999 Delivered-To: freebsd-questions@freebsd.org Received: from web1305.mail.yahoo.com (web1305.mail.yahoo.com [128.11.23.155]) by hub.freebsd.org (Postfix) with SMTP id 8363A14C37 for ; Sat, 7 Aug 1999 12:28:27 -0700 (PDT) (envelope-from dipierro5@yahoo.com) Message-ID: <19990807193705.17890.rocketmail@web1305.mail.yahoo.com> Received: from [209.71.85.198] by web1305.mail.yahoo.com; Sat, 07 Aug 1999 12:37:05 PDT Date: Sat, 7 Aug 1999 12:37:05 -0700 (PDT) From: Anthony DiPierro Subject: Re: maximum TCP connections To: roelof@nisser.com Cc: freebsd-questions@freebsd.org MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: owner-freebsd-questions@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Anthony DiPierro wrote: > > What is the overhead of a TCP connection on freebsd? > I need to support "lots" (how many is the question) of > connections to multiple different locations, while > noting disconnections (both graceful and ungraceful) > and connections. I'd love to make the limiting factor > the amount of traffic (which will be low compared to > the number of connections), and I'm trying to figure > out if I should use generic TCP, UDP, or maybe a > hacked up version of TCP. > > Feel free to tell me to RTFM, if you'd be so kind as > to tell me where the FM is. Not that it would allow me to give the answer but I think that the most important piece of data would be the available bandwith. Are we talking fiber here or a T1? Also take a look at the Walnut Creek site. They run FreeBSD and hold the current world record in daily transfers. Maybe their specs could tell you something. Roelof PS the URL is http://www.cdrom.com/ ------------------------------------------------- Thanks for the response. Getting bandwidth to the machine is not a factor. We're going to be colocated at a site that can provide us with (and charge us for) all the bandwidth we can handle. Now if there is some basic bandwidth overhead with TCP (pings to check for a maintained connection, for instance), that would be important to know. But the bandwidth for the actual application would be much less per connection compared to cdrom.com. I would basically be sending short status messages back and forth every few minutes on average. My hope is that this bandwidth will become the limiting factor in how many connections I can allow per machine. I could definately make this the limiting factor with UDP (assuming I throw enough ram in the machine to hold each connection status entry, 1 gig / 100 bytes = 10,000,000 connections), I'm trying to see if there is some other limit in the TCP overhead per connection which would force me to use this UDP solution rather than just letting the kernel handle the reliability and connection maintanence aspects for me. I couldn't find the stats on the cdrom.com machine. How many connections they handle in addition to the specs of the machine(s) would definately be a good starting place. I'm probably going to have to wind up looking at the kernel source to see what's going on. I'll probably need to look at it to tweak some tunables anyway though. _____________________________________________________________ Do You Yahoo!? Bid and sell for free at http://auctions.yahoo.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message