From owner-freebsd-questions Mon Jul 8 03:12:13 1996 Return-Path: owner-questions Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id DAA22792 for questions-outgoing; Mon, 8 Jul 1996 03:12:13 -0700 (PDT) Received: from aries.ai.net ([208.194.41.4]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id DAA22784 for ; Mon, 8 Jul 1996 03:12:09 -0700 (PDT) Received: (from nc@localhost) by aries.ai.net (8.6.11/8.6.12) id GAA28426; Mon, 8 Jul 1996 06:11:59 -0400 Date: Mon, 8 Jul 1996 06:11:58 -0400 (EDT) From: Network Coordinator To: Jaye Mathisen cc: David Greenman , freebsd-questions@freebsd.org Subject: Re: Increasing FTP thruput. In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-questions@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Whoah. I tried the same thing on an NT box and my jaw dropped. 4.76Kbytes/second on the same 3MB file. Amazing. I wouldn't even mind having two separate FTP sites on the same machine [with different optimizations on each]. I would, however, like to find out what specifically NT is doing differently than FreeBSD 2.1-stable. Thanks for your information! -Tomas On Sun, 7 Jul 1996, Jaye Mathisen wrote: > > All this might be true, but the article I sent a reference out about > recently pointed out several shortcomings in TCP re slow speed links, and > that tweaks that helped the high end were having a negative effect on the > low end. > > Sun specifically rewrote the drivers and tweaked some constants to help > with this specific issue, so it doesn't surprise me at all. > > My users see similar behaviour. They can FTP files faster from an > unloaded NT box than they can from an unloaded FreeBSD box. yet both can > drive the thernet at wire speeds. (They're coming in over a 28.8 PPP > connections to Livingston PM's and Digiboards) In both cases the configs > are almost identical, P120's, 3com ethernet, 2GB Seagate disks, NT3.51 on > a Buslogic, and FreeBSd on and adaptec. > > This is reproducible at will. > > So now after I went to all the hassle to switch from BSD/OS to FreeBSD, I > find I probably would be able to keep more users happier by changing > things to NT. This is depressing. > > I have the original article still, if anybody is interested. > > On Sun, 7 Jul 1996, David Greenman wrote: > > > >On one of our development machines we are running wu-ftpd. When FTP'ing > > >over a T-3 to a PPP (28.8kbaud) line from a PPro 200 to it we were seeing > > >as low as 2.0k/second. Yet, FTP'ing from a Sun 4/75 we were seeing > > >2.8kbaud solid performance. > > > > > >It appears to be [by watching modem lights] that the BSD box isn't > > >streaming in the sense of a continuous send or receive while receiving > > >acks along the way. The Sun boxes [all of them] seem to do this out of > > >the box. > > > > > >I verified this behavior on ftp.cdrom.com and random SunOS boxes. > > > > > >Any ideas how to make the BSD behavior more like the SunOS? > > > > Watching modem lights is an extremely poor way to do critical analysis. The > > SunOS machines send out ACKs every other packet and further restrict the > > window size to 4096 bytes. I'm really suprised that FreeBSD comes out worse > > in your test. If you're doing the tests using ftp.cdrom.com as the endpoint > > then you might be seeing the effects of asymetric lossage - especially since > > there are so many intermediate hops involved. I suggest looking for dropped > > packets in the netstat -s statistics. > > Oh, and you didn't mention which version of FreeBSD you're using or whether > > you're using the kernel PPP or usermode ijppp. You could be seeing the effects > > of one or more bugs in the PPP code. > > > > -DG > > > > David Greenman > > Core-team/Principal Architect, The FreeBSD Project > > > >