Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 18 Nov 2004 22:29:17 +0100
From:      Michael Schuh <michael.schuh@gmail.com>
To:        freebsd-stable@freebsd.org
Subject:   serious networking (em) performance (ggate and NFS)/Performance issues
Message-ID:  <1dbad31504111813293da2d798@mail.gmail.com>
In-Reply-To: <1dbad31504111813135e011a46@mail.gmail.com>
References:  <20041118003500.946B916A50A@hub.freebsd.org> <1dbad31504111813135e011a46@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
From: Michael Schuh <michael.schuh@gmail.com>
Date: Thu, 18 Nov 2004 22:13:00 +0100
Subject: Re: serious networking (em) performance (ggate and
NFS)/Performance issues
To: freebsd-stable-request@freebsd.org


Hi,

i have same serious problems.
I don't repeate the posting, and describe my tests that i have made
with RELENG_5 (Original Release)  and RELENG_4

My Hardware scenario:
Server
PIII 866MHz
512MB RAM
OS-Disk
SCSI- DNES IBM
3*UDMA66-Drives as Concat with ~140G
Rsync with SO_KEEPALIVE
3COMXL (3C905-Combo)

Client
AMD K6-2 450MHz
384 MB RAM
UDMA33-Disks
3COMXL (xl,all models)
DEC21143 (Tulip, dc)
RTL8139(rl)

The only differences between the Tests are the OS_versions.
In Both OS-Versions all the Binaries and the Kernel are compiled on the
Client. For RELENG_4 CPUTYPE=i586, RELENG_5 CPUTYPE is overwritten in
/etc/make.conf as k6-2. Both kernels are GENERIC except the following Options:

cpu i586 #only
options HZ=2000
options DEVICE_POLLING

The Serverside was constant RELENG_4 with same options.
On the Server i has pure DISK I/o-Rates from ~20 MB for wirte and
~30MB for read, so that these not could be the bottleneck.

The only bottleneck in the system are the capability of max working interrupts
and physically the Network. The Boxes are connectet with daetwyler
cat. 5e through
an 100MBit exclusive switch that would only be used from these two boxes.

In all the Tests i have made only a Push from Client to server, while
i have seen that the
Server is not the Problem. I have alwas used the same Data to transfer
(gziped dump -0 from
my systems ca 5.3G)
All the tests are made with only one of the described networkadapters
on the same Time,
the problems seems not to be on the driver side.Maybe?

The interruptrates and transfers disk I/Os:
RELENG_4
Client maxinterrupt ~11000/s
Client maxpackages ~6000/s
Client Disk i/o read    ~7-8MB/s
Transferrates  ~7-8MB/s similar to Disktraffic

RELENG_5
Cleint maxinterrupt ~7000/s
Client maxpackages ~2500-3000/s
Client Disk i/o read ~3-4MB/s
Transferrates ~3-4MB similar to Disktraffic

Is now the Diskdriver the Problem or the networklayer?
Who has also these expieriences?

This made me a little bit ugly, and i think i should be conservative
and stay on RELENG_4. Or is this the price for the "new technologie features"?
In the first moment i have thinking i have a linux box and not FreeBSD.

Can the Developers made us happy with the good old high performance from
RELENG_4 on RELENG_5??

we hope the best.

best regards
sicerelly
michael



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1dbad31504111813293da2d798>