*BSD News Article 9348


Return to BSD News archive

Received: by minnie.vk1xwt.ampr.org with NNTP
	id AA5614 ; Fri, 01 Jan 93 01:50:55 EST
Path: sserve!manuel.anu.edu.au!munnari.oz.au!spool.mu.edu!agate!ames!pacbell.com!hoptoad!curt
From: curt@hoptoad.uucp (Curt Mayer)
Newsgroups: comp.unix.bsd
Subject: Re: NE2000 test results (previous driver performs better). NOT!
Message-ID: <30224@hoptoad.uucp>
Date: 27 Dec 92 22:27:12 GMT
References: <1992Dec27.181634.523@ponds.uucp>
Organization: foobie bletch
Lines: 79
Keywords: ether, speed, 386bsd NE2000

In article <1992Dec27.181634.523@ponds.uucp> rivers@ponds.uucp (Thomas David Rivers) writes:
>
> Well, since Curt Mayer (curt@hoptoad.uucp) had re-written the
> NE2000 driver for us, I decided to take some time and give it
> a good shaking out.
>
>     ponds			lakes
>   386DX, 33mhz	     386DX, 33mhz
>   NE2000 clone             NE2000 clone
>   8-meg memory             12-meg memory
>   Hercules		     ET4000 clone (running XFree86 1.1)
>   IDE 240Meg.		     SCSI -663 meg
>   Wangtek (EN2099) 	     Wangtek (5150ES)
>   40-meg swap on wd0	     40-meg swap on sd0
>
>   Time to Transfer (seconds)  Kbytes/Sec
>     4.3                        1e+02
>     3.3                        1.3e+02
>     3.4            		1.3e+02
>     2.8			1.6e+02
>     4.2			1e+02
>
> These appear to be rather good time, approaching what you would
> expect the bandwidth to be on such a network (few collisions...)
>

like hell. the new driver's best speed running ftp was 621 kb/sec
on a ne2000 clone. the sustained speed (measured over 20 minutes)
was 456kb/sec. this is on a UMC 486/50 motherboard running a 10mhz
isa bus.

note that the old driver's performance on my hardware is consistent
with thomas' results above.

also, try nfs read performance at 620 kb/sec

> Ok, so I installed the driver posted by Curt.  My original intent
> was to install it on both machines, then only one. However, when 
> that kernel came up on "lakes" the network didn't function (lots of
> overun-type errors from the driver: ne0 ... error: isr 15; 
> ne0 ... error: isr 14, etc...) 
>
> Thus, I was only able to get Curt's driver to run on "ponds".  The 
> following times reflect that situation (ponds is running Curt Mayer's
> driver, lakes is running the driver posted in July.)  There was *no*
> other change in the test, hardware, etc...  Since the times appeared
> so much worse, I ran more tests than before.
>
>   Time to Transfer (seconds)  Kbytes/Sec
>     10				42 
>     10				42
>     11				39
>     10				42
>     15				29
>     6.3			69
>     9.4			46
>
> These numbers don't approach those of the other driver.  Also, just to
> get a feel for NFS mounts, rlogin, etc... I tried out several NFS
> configurations, rlogin'ing and so forth - generally it seemed to be
> much slower.
>
> OK, so I thought I may have seriously goofed things up, since these
> numbers were so much lower than the other driver.  So, I rebooted
> ponds with the (saved) kernel containing the Friedl drivers, and
> ran some tests.  Again, the only change I made was to reboot the
> machine - no other modifications.  The test results returned to a set
> similar to above, indicating that it was indeed the NE2000 driver
> which explained the slowdown.
> 
I'm investigating why your performance was so much worse.
but 100k/sec is not even close to acceptable on unloaded ether.

>        - Dave Rivers -
>        (rivers@ponds.uucp)
-- 
	curt mayer
        curt@toad.com
        415-387-0217 home