Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 17 Sep 1997 10:29:53 +0930
From:      Greg Lehey <grog@lemis.com>
To:        Jan Martin Borgersen <borg@cs.unc.edu>
Cc:        freebsd-hardware@FreeBSD.ORG
Subject:   Re: 10Mbit Ethernet Adapter Performance
Message-ID:  <19970917102953.15249@lemis.com>
In-Reply-To: <199709170018.UAA19721@buzzard.cs.unc.edu>; from Jan Martin Borgersen on Tue, Sep 16, 1997 at 08:18:45PM -0400
References:  <199709170018.UAA19721@buzzard.cs.unc.edu>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Sep 16, 1997 at 08:18:45PM -0400, Jan Martin Borgersen wrote:
>
> Can anyone point me to any performance evaluations
> of 10Mb ISA Ethernet adapters or their device drivers?
>
> I'm getting some really interesting numbers:
>
> On an isolated, 100Mb switched Ethernet hub with no
> other network traffic, I'm running ttcp to perform
> tcp and udp blasts between several 486-66 class
> machines with ISA 10Mb Ethernet adapters.
>
> When machines with generic IBM cards (looks like
> a NS 83905A chip) thrown into NE2000 emulation and
> using the ed0 driver are transmitting, I'm seeing
> network throughputs on the order of 8300 bps.

I hope these are Bytes per second, and not bits per second.  Not that
it makes much difference with those rates.

> When machines with 3Com 3C509 Etherlink III's using
> the ep0 driver are transmitting, I'm seeing network
> throughputs on the order of 3500 bps.

You could always install a modem...

> Interestingly, the machine I have with a PCI 3C590
> 3Com card using the vx0 driver is transmitting
> at about 7700 bps -- still significantly SLOWER than
> the NE2000 emulated cards!!!!
>
> I've swapped cards among machines -- the bad numbers
> follow the 3C509's and ep0 drivers, and seem to be
> much more a function of transmitter than receiver.
>
> What happened to FreeBSD's 3Com drivers?????  (Or am
> I not configuring something for optimization?)

You can be pretty sure that this is a network configuration problem of
some nature.  Some boards are faster than others, of course, but using
reasonably fast machines, an ftp transfer should give you throughputs
in the order of 1 MB/s.  Here's an example of transfers I did just now
between a 486DX/2-66 with a WD8013EPC and a P133 with a 3C509:

1242318 bytes received in 2.39 seconds (507.83 KB/s)
1242318 bytes received in 1.26 seconds (963.66 KB/s)
1242318 bytes received in 1.17 seconds (1.01 MB/s)
1242318 bytes received in 1.12 seconds (1.06 MB/s)
1242318 bytes received in 1.17 seconds (1.01 MB/s)

You can assume the slow transfer time in the first attempt was due to
the slowness of the 486 machine reading in the file; after that, the
data was in cache, so we're measuring the transfer speed.

So, what's wrong with your network?  Check:

1.  The output of netstat -in.  It should look something like:

    Name  Mtu   Network       Address            Ipkts Ierrs    Opkts Oerrs  Coll
    ep0   1500  <Link>      00.a0.24.37.0d.2b   231549     2   244717     4     0
    ep0   1500  255.255.255&0 192.109.197.137   231549     2   244717     4     0

    If the network is idle, you shouldn't have any collisions.  Even
    on a heavily loaded network, you'll be in trouble with more than
    1% collisions.

2.  What's going on on the net?  Use tcpdump to see if you're getting
    lots of retransmissions.

3.  Normally, I'd be inclined to say "check the boards", but if you're
    getting the same performance out of a number of different boards,
    they're unlikely to be the problem.

In any case, the problem isn't FreeBSD.  If you look at your net,
you'll probably find something.

Greg




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970917102953.15249>