[Prism54-users] Prism54 performance revisited ...

Luis R. Rodriguez mcgrof@ruslug.rutgers.edu
Thu, 4 Dec 2003 12:23:26 -0500 (EST)


Wow. Cool ;) this is going to our NEWS section :-D
	Luis

On Thu, 4 Dec 2003, Stefan U. Hegner (Mobil) wrote:

> Hi there,
>
> just did a couple of performance benchmarks with the 20031130-cvs driver
> and the results are very good! - Developers: please be encouraged!
>
> What I did:
> I copied a 200MB iso-image using scp over the air. I did this for Linux
> as well as for Windows (using winscp2). Here are the results:
>
> Prism54 card:	SMC2835W
> Access point:	U.S.Robotics USR 808054
> Wired client:	Dual P3-500 running Gentoo 1.4 and Openssh
> Link strength (from same position): -60db (Linux) 60% (Windows)
> Link distance is about 8 mtrs. - There is one brick wall or two glass
> doors between the AP and teh Prism54 card.
>
> Linux Throughput
> scp from Prism54 to AP:	1.1 MB/s      net bandwidth: 8.8Mbit
> scp from AP to Prism54:	1.9 - 2.3MB/s net bandwidth: 15.2 up to 18.4Mbit
>
> I got a stable 2.2 MB/s link translating to 17.4MBit. I think this a
> cool bandwidth! - However, I don't understand why there is such a
> difference in sending and receiving data.
>
> Windows Throughput
> scp from Prism54 to AP:	0.9 MB/s      net bandwidth: 7.2Mbit
> scp from AP to Prism54:	0.6 MB/s      net bandwidth: 4.8Mbit
>
> However, the last test result seems very strange to me. And I cannot
> believe that WinSCP is creating to much overhead on my P3-1200 notebook,
> that io drops.
>
> Anyhow, I think this is a more accurate measure than the bing test. And
> since ssh/scp is on most Linux machines, this can be easily tested. The
> stated MB/s are an estimated average from the Throughput displayed
> during the copy process.
>
> HTH.
>
> Stefan.
>
> --
> / Stefan Ulrich Hegner aka "Hegi" - Loehne/ Westfalen - Germany \
> | mobil@hegner-online.de - primary Email stefan@hegner-online.de|
> \ GPG-Fprint D9DB 51BD 2DA6 9B3A 41CB  0287 05A1 8D11 38BA CE91 /
>