Date: Wed, 10 Dec 2003 23:35:09 -0000 From: Alex (ander Sendzimir) <xela@battleface.com> To: freebsd-questions@freebsd.org Subject: network throughput and kernel configuration... Message-ID: <7CB01C96-2B69-11D8-9215-000A95775140@battleface.com>
next in thread | raw e-mail | index | archive | help
I've been experimenting with kernel configurations and network
performance. I'm running a dual Xeon processor (500MHz) system with
10/100 ethernet (Intel). It's an Intel motherboard. The cable is CAT5
and I'm transferring between the FBSD box and an Apple G4 laptop with
10/100 ethernet. This is going through a Netgear 10/100 hub (non
switching). In my limited capacity I'm using ncftp{put/get} to transfer
large files. Using a basic, no frills custom kernel (nothing
controversial), ncftp shows about 7.5MB/s throughput between these two
machines. I added the following lines to my kernel conf
options DEVICE_POLLING
options HZ=1000
and modified /etc/sysctl.conf by adding
kern.polling.enable=1
and now ncftp shows about 8 MB/s. (Recompile, reboot, etc)
Perhaps I'm dreaming, but shouldn't I be getting better network
throughput in either case? Perhaps a switch would be better than a hub?
The G4 Powerbook shows its running en0 at half-duplex over 100baseTX.
Which, I think, is the same for the dual Xeon system. Both are
autoselect. Perhaps I should scrap the kernel config issue and
configure both interfaces for full-duplex 100baseTX.
Thoughts/experience/advice?
Thanks for the help, as always.
Alex
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Alexander Sendzimir 802 863 5502
Mac Tutor of Vermont, LLC info @ mactutor . vt . us
Colchester, VT 05446
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?7CB01C96-2B69-11D8-9215-000A95775140>