Date: Wed, 21 Oct 1998 08:30:00 -0700 From: Tamiji Homma <thomma@BayNetworks.COM> To: matthew@wolfepub.com Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: Network device slowdown? Message-ID: <19981021083000X.thomma@baynetworks.com> In-Reply-To: Your message of "Tue, 20 Oct 1998 22:46:56 -0400" <3.0.5.32.19981020224656.007b76b0@firebat.wolfepub.com> References: <3.0.5.32.19981020224656.007b76b0@firebat.wolfepub.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> I was wondering if there are any known problems with the ed driver? I had > a Kingston 10/100 PCI ethernet card based on a DEC chipset in my 2.2.6-R > box and the performance was terrible (<1Mbs). I switched to a plain ISA > 10Mbs ethernet card and the performance shot through the roof. I figured > that the driver for DEC based cards was probably not mature yet, so I stuck > with the ISA ethernet. Coincidentally, I was wondering(how many more?) if there is a way to improve(or fix) de0 (sorry not ed driver) performance. The problem that I have is: Machine A: Machine B: FIC PA-2007 K6/233 PR-440FX P6/200-512KBx2 SMP 64MB EDO 128MB EDO ECC -current(10/17/98) -current(10/17/98) Quantum Fireball 6.4G IBM 9ZX (9G) Maxtor 2880 11.5G IBM 9ES (4.3G) IDE DMA is on. I used the following command to get timing (I know it is not a good one but...) # rsh machineA dd if=/dev/zero bs=64k count=1000 >/dev/null (on machine B) # rsh machineA dd if=/dev/zero bs=64k count=1000 >/usr/tmp/junk (on machine B) # rsh machineB dd if=/dev/zero bs=64k count=1000 >/dev/null (on machine A) # rsh machineB dd if=/dev/zero bs=64k count=1000 >/usr/tmp/junk (on machine A) A ---> B means that a large amount of TCP transfer from machine A to B. Machine A and machine B are connected by cross-over cable for this experiement. 100Mbps/Full-duplex: Machine A Machine B de0 ----> fxp0 (over 9MByte/sec with disk write) de0 <---- fxp0 (somewhere 60KB/sec with disk write) fxp0 ----> fxp0 (11MB/sec without disk write) fxp0 ----> fxp0 (9MB/sec with disk write) fxp0 <---- fxp0 (9MB/sec without disk write) de0 ----> de0 (over 7MB/sec without disk write) 10Mbps/Full-duplex: de0 <---> de0 (1.06MB with disk write) netstat -i shows that de0 on FIC-2007 is having 2 to 3 ierrors per second when de0 receives a large amount of data with 100Mbps. That explains terrible performance 60KB/sec. fxp0 sees no errors on both machines as well as de0 on PR-440FX has no errors. When I switch to 10Mbps, there is no error what so ever on both de0 and fxp0 on both machines. The experiment above is not really systematic yet but one thing I can tell is that de0 on FIC PA-2007 (VP2/97 chipset) does not handle 100Mbps input stream very well but 440FX does it well with both fxp0 and de0. I tried tweaking PCI timing to be all aggressive on FIC PA-2007 but it doesn't show any visible effect. Well, it is very hard to know what is really changed in PCI bus timing parameters from BIOS menu :( Just data point... Well, next step is to read de driver code to see if anything I can do with it. Tammy To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19981021083000X.thomma>