Date: Tue, 01 Jan 2002 17:22:35 -0800 From: "Kevin Oberman" <oberman@es.net> To: gregory.lane@anu.edu.au Cc: freebsd-questions@FreeBSD.ORG Subject: Re: Pathetic 11 kbytes/s with ed driver (Netgear EA201) Message-ID: <200201020122.g021MZd28758@ptavv.es.net> In-Reply-To: Your message of "Wed, 02 Jan 2002 11:56:48 %2B1100." <20020102115648.A35606@nucl03.anu.edu.au>
next in thread | previous in thread | raw e-mail | index | archive | help
> Date: Wed, 2 Jan 2002 11:56:48 +1100 > From: Greg Lane <gregory.lane@anu.edu.au> > Sender: owner-freebsd-questions@FreeBSD.ORG > > I am putting together a cheap webserver out of a discarded 486 with > all ISA slots. I am using a Netgear EA201 ISA network card which > uses the ed driver. I get CRAP bandwidth out of the thing. Using > fetch to get a file: > > 3>xxxxx@router:~$ fetch http://rene:/mysql_manual.pdf > Receiving mysql_manual.pdf (3314090 bytes): 100% > 3314090 bytes transferred in 279.0 seconds (11.60 kBps) > > This is on a local network where the client and server are > each plugged into a 10baseT hub. > > If I try using scp to copy a file, I can get better, sometimes > 100kBps. There is definitely some problem though as the traffic > shows pauses (see below in a tcpdump). First the config info: > > I have used the Netgear DOS configuration program to set the > card to non-PNP mode, use IRQ 5, port 0x300 and full-duplex. Are you sure that this card supports full-duplex? If you are plugged into a 10Base-T hub (and not a switch), your MUST run half-duplex. What does ifconfig ed0 show? > It is recognised fine (excerpt from dmesg): > > ed0 at port 0x300-0x31f iomem 0xd8000 irq 5 drq 0 on isa0 > ed0: address 00:40:05:a2:15:fc, type NE2000 (16 bit) > > My kernel config has the line > > device ed0 at isa? port 0x300 irq 5 iomem 0xd8000 > > One question? Is the iomem used? There is no way to set this > that I saw with the Netgear DOS configuration program. I read > somewhere that if the shared memory is not properly set up > then it can cause troubles, but I don't know whether shared > memory is set with the iomem setting. It is used and should be OK as the ORM device should prevent it from overlapping with anything else. > Tcpdump output during the fetch is at the very bottom of this > mail. Notice the one second long gaps. > This is obviously the problem but why does it occur? My first > guess was a mismatch in speed and full/half duplex. It is duplex. If it was a speed mis-match, you would not get ANY data through. One second gaps indicate that TCP data is sent, but the ACK is not received and TCP times out and re-transmits. If you have a half/full duplex mis-match it is likely that your third packet is colliding with the ACK of the second so that the second is not ACKed and the third is not received by the other node. Everyone steps back for about a seconds and starts all over. > I have not tried a crossover cable because these boxes are all > headless and I could not get remote access if I did this. A cross-over cable will not help. Like speed, if it is wrong, nothing will work. > Lastly, here is netstat output showing there are no collisions. This only confirms that you are running full-duplex. In full-duplex there are no collisions because that is what full-duplex really means. Don't detect collisions and don't wait for the media to be free before transmission. Make sure that ALL devices connected to a 10Base-T hub are set to half-duplex. R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200201020122.g021MZd28758>