[rhn-users] Gigabit transfer speeds

[Tim Schoenfelder]

Increase your packet size...

One of our networking class projects a couple years ago was to benchmark
gigibit ethernet transfer speeds.

I benchmarked 700mb/s+ between two Linux PCs which was reasonable to our
instructor.

He emphasized that the brand/chipset of cards made a big difference ( it
wasn't the most expensive that performed best ... ) as well as the drivers.

Some of his published tests as described below show that you can achieve at
least 700mb/s even with 32bit Netgear GA622T cards as illustrated in a ps
link from his website:

http://www.cs.uni.edu/~gray/

Benchmark project links:
http://www.cs.uni.edu/~gray/gig-over-copper/
http://www.cs.uni.edu/~gray/gig-over-copper/gig-over-copper.html

PS file showing performance benchmarks ( note 700mb/s+ requires packet size
of 100k+ bytes ):

http://www.cs.uni.edu/~gray/gig-over-copper/hsln-lcn.ps

Tim

On 8/21/06, Corne Beerse <cbeerse at gmail.com> wrote:
>
> Dan Hunter wrote:
>
> > I regularly move large files from a windows XP machine to a Fedora
> > Core 5 machine.
> > To avoid bothering other network users, I purchased two Linksys EG1032
> > gigabit ether net cards.
> > The cards have been installed.  a direct PC to PC gigabit cross over
> > cable connects them.
> > A separate set of IP address have been set up.
> > I can verify that data is being transfered over the gigabit cable with
> > windows task manager.
> >
> > Transfers over the 10/100 link show up at 82 or 83 percent - about 80
> > Mbps.
> > Transfers over the gigabit link show up at 20 to 26 percent - about
> > 230 Mbps.
> >
> > What do I have set up wrong?
> >
> >
> Your expectations are to high. If you have about 20% utilisation with a
> Gbit nic, I'd say you are doing verry nice.
>
> With the move to gigabit networking, the nic is no longer the
> bottleneck. If you move data from disk on one machine or to disk on an
> other, the disk-controllers, disk interfaces or even the actual disks
> are most likely the bottlenecks. Then, the disk does not need to be the
> bottleneck. Think about the speed of internal busses and for which they
> are used while you are dumping data...
>
>
> CBee
>
> _______________________________________________
> rhn-users mailing list
> rhn-users at redhat.com
> https://www.redhat.com/mailman/listinfo/rhn-users
>



-- 
Tim Schoenfelder
http://timschoenfelder.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/rhn-users/attachments/20060821/223735e8/attachment.htm>