From owner-freebsd-net Wed Mar 3 1:38: 8 1999 Delivered-To: freebsd-net@freebsd.org Received: from ns11.nokia.com (ns11.nokia.com [131.228.6.230]) by hub.freebsd.org (Postfix) with ESMTP id 093F214BE7 for ; Wed, 3 Mar 1999 01:38:05 -0800 (PST) (envelope-from yu.shi@research.nokia.com) Received: from pepper.research.nokia.com (pepper.research.nokia.com [131.228.12.3]) by ns11.nokia.com (8.8.8/8.6.9) with ESMTP id LAA17170; Wed, 3 Mar 1999 11:34:25 +0200 (EET) Received: from pupu.research.nokia.com (pupu.research.nokia.com [131.228.13.130]) by pepper.research.nokia.com (8.9.1a/8.9.1) with ESMTP id LAA20341; Wed, 3 Mar 1999 11:34:23 +0200 (EET) Received: from research.nokia.com ([172.28.31.90]) by pupu.research.nokia.com (8.9.1a/8.9.1) with ESMTP id LAA07455; Wed, 3 Mar 1999 11:31:48 +0200 (EET) Message-ID: <36DD0289.7B2BC067@research.nokia.com> Date: Wed, 03 Mar 1999 17:36:09 +0800 From: Shi Yu Reply-To: yu.shi@research.nokia.com Organization: Nokia China X-Mailer: Mozilla 4.02 [en] (Win95; I) MIME-Version: 1.0 To: EXT Chris Williams Cc: freebsd-net@FreeBSD.ORG Subject: Re: TCP packet sizes and window w/X through SSH References: <36CDC4E7.CC332C19@geekspace.com> Content-Type: text/plain; charset=gb2312 Content-Transfer-Encoding: 7bit Content-Transfer-Encoding: 7bit Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org For question1, I think it is up to the application to increase the receive and send buffer size by using function 'setsockopt'. For question2, small packets may be generated by writing a small mount of bytes to socket and disabling Nagle algorithm. EXT Chris Williams wrote: > 1) During peak data transmission between the two unix boxes, both sides > frequently hit the maximum window size before getting an ACK, and then > have to wait for the ACK to send more. Since there are no > retransmissions occuring (woo!), it would seem that an increase in > window size should boost performance. Is there any reasonable way to do > this in FreeBSD? I looked through sysctl -a and didn't see anything, > 'man tcp' isn't helpful...Is it an option I have to set before compling > the kernel? Or what? > > 2) The SSH packets going from the BSD box to the proxy server are almost > always less than ~800 bytes. This seems strange since the Ethernet MTU > is 1500, and most of the packets coming from the Sun box to the BSD box > to be relayed through SSH are in fact ~1500 bytes. Why is SSH chopping > the packets up like this? Is there a good reason for it, and if not, how > can it be fixed/changed? > Although this doesn't appear to be slowing things down as much as (1), > it does put extra load on our proxy (from watching perfmon against the > proxy while I do varios things, it is very clear that cpu/io load is > much more closely coupled with packets/sec than bytes/sec -- which isn't > at all surprising). > > Any suggestions on either point would be much appreciated.. > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-net" in the body of the message To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message