Date: Thu, 10 Nov 2005 13:24:24 +0000 (GMT) From: Robert Watson <rwatson@FreeBSD.org> To: net@FreeBSD.org Subject: TCP inflight (was: Re: Poor Samba throughput on 6.0 (fwd)) Message-ID: <20051110132309.A68007@fledge.watson.org>
next in thread | raw e-mail | index | archive | help
I've seen a number of reports that TCP inflight data limiting is
substantially impacting performance on high bandwidth, low latency
networks. Not knowing much about it, I figured I'd post on net@ and see
if anyone here is interested in taking a look at some of the reports
(they're easy to find in the mailing list archives) and investigating how
to fix them.
Robert N M Watson
---------- Forwarded message ----------
Date: Thu, 10 Nov 2005 10:31:03 +0000
From: Joao Barros <joao.barros@gmail.com>
To: Robert Watson <rwatson@freebsd.org>
Cc: Mike Jakubik <mikej@rogers.com>, freebsd-performance@freebsd.org,
Arkadi Shishlov <arkadi@mebius.lv>
Subject: Re: Poor Samba throughput on 6.0
On 11/10/05, Robert Watson <rwatson@freebsd.org> wrote:
> On Wed, 9 Nov 2005, Mike Jakubik wrote:
>
>> I have done many tests to try to determine the poor performance on my
>> systems (FreeBSD-current connected directly to Windows XP via identical
>> Intel Pro 1000 cards) and my only conclusion is that Samba on FreeBSD
>> when talking to a Windows box is simply fubared. Im willing to bet thats
>> not the case on Linux. I can get great speeds (~30MB/s) using ftp
>> between the machines, but when i use Samba, i get a maximum of 11MB/s. I
>> have tried playing with mtu size, samba options, polling, but there does
>> not seem to be any visible bottlenecks, its just simply slow. It would
>> be great if someone could get to the bottom of this, its quite
>> frustrating.
>
> There are a number of TCP related configuration frobs on FreeBSD. It
> would be quite interesting to know how modifying each of the following
> settings affects Samba performance:
>
> net.inet.tcp.delayed_ack
> net.inet.tcp.sack.enable
> net.inet.tcp.inflight.enable
>
> There has been recent evidence that the inflight bandwidth calculation is
> having problems generating stable performance at high bandwidth and low
> latency, so I might try that one first.
>
> Robert N M Watson
I just tested on my home machine and delayed_ack and sack.enable have
little or no effect on the results.
inflight.enable on the other hand had a huge impact! With the file
cached I've gone from flat 5.5MB/s up to also flat 7.2MB/s
It's an improvement but there is still a difference up to those
theoretical 12.5MB/s on a 100mbit link
--
Joao Barros
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20051110132309.A68007>