Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 29 Apr 2019 14:08:25 -0600
From:      Alan Somers <asomers@freebsd.org>
To:        Eugene Grosbein <eugen@grosbein.net>
Cc:        Wojciech Puchar <wojtek@puchar.net>,  "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org>
Subject:   Re: very low performance tcp/rsh
Message-ID:  <CAOtMX2hAeq%2B7GwdXu9MptrMk%2BY%2B37zebTU%2Bn41j0NT5-UVz53Q@mail.gmail.com>
In-Reply-To: <cb76aa41-1472-5df1-5b24-384f6453f3de@grosbein.net>
References:  <alpine.BSF.2.20.1904292028400.48623@puchar.net> <ef440d5f-c543-148a-4e68-44b0bc62b28f@grosbein.net> <alpine.BSF.2.20.1904292111080.57162@puchar.net> <cb76aa41-1472-5df1-5b24-384f6453f3de@grosbein.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Apr 29, 2019 at 1:57 PM Eugene Grosbein <eugen@grosbein.net> wrote:
>
> 30.04.2019 2:22, Wojciech Puchar wrote:
>
> > create 32GB hole-file
> >
> > dd if=/dev/zero of=test bs=32k seek=1m count=0
> >
> > fetch -o /dev/null over ftp (proftpd) gives 1.4GB/s
> >
> > proftpd consumed 100% of single core. much better.
> >
> > tested with FreeBSD ftpd  - 1.7GB/s
> >
> > seems like freebsd can saturate 10Gb/s ethernet with single core (Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz)
> >
> >
> > Still where is the most overhead? one memcopy to packet buffers+creating packet headers (send).
> >
> > reception is on separate process taking another core.
> >
> >
> > still - quite good. But compared to filesystem overhead - large.
>
> Do not use file with holes (holes require special processing taking lots of CPU cycles).
> Create ordinary file within RAM disk and use sock ftp daemon for sake of sendfile().

Better yet, if the goal is simply to measure TCP performance, use
benchmarks/netperf.  Don't try to do anything involving files.

-Alan



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAOtMX2hAeq%2B7GwdXu9MptrMk%2BY%2B37zebTU%2Bn41j0NT5-UVz53Q>