[vortex] Performance tests with 3c905B

Thomas Ackermann tgm@math.uni-bonn.de
Sun, 23 Sep 2001 16:59:41 +0200


On Fri, Sep 21, 2001 at 01:50:41PM +0200, Bogdan Costescu wrote:
> > I tested with that commandm reading file from the filesystem and coping to
> > /dev/null on the destination host:
> 
> This is usually not a good way to measure network speed, as you include
> other parameters, like disk speed, cache performance, etc. which might be
> bottlenecks. Usually, network performance is best measured using tools
> like ttcp or netperf which are network specific (send data from their own
> buffer(s) directly using TCP or UDP, thus not involving any disk/cache).

Thanx - i don't know that tools. I use "bing" for such things normaly and
suppose that it is similar to yours ...
I esp. try to read from the filesystem and over caches because i want to
make backups over the net.


> > By the way, what do you think or know, is the *best* 100Mb PCI-NIC for
> > Linux - regarding sheer performce (or maybe what's the best such NIC for
> > a server and for a client)?
> 
> Heh, you are talking about 3c905B and asking for a better card ? 8-)
> 
> On the beowulf (for which network performance is a must) mailing list,
> people seem to agree that Intel cards (driven by eepro100/e100), 3Com
> cards (driven by 3c59x) or DEC 21040 and derived cards (driven by tulip)
> give good results; also that Realtek 8139 (driven by rtl8139/8139too)
> should be avoided.

That's an important information! But the driver by itself does not give 
info about the quality of a card, does it?
I mean, only because a card is driven by 3c59x does not mean that the card
realy is good, or?
Can you tell me some explizit cards that are known as good?
(For example 3c905B compared to 3c905C ...)

Thanx for your reply and
	thanx for any further information	;-)
	Byebye,
-- 
        Thomas Ackermann | Tel. +49-(0)228/631369 | Mobil: 0178-2016033
                finger tgm@rhein.math.uni-bonn.de for public key
                         Email: <tgm@math.uni-bonn.de>