Very Computer

On a machine running 4.5-RELEASE, and the network was just too
intolerably slow; on a 10baseT LAN, transfer rates averaged 4K/sec.
This seemed consistent with X, FTP, and just about everything else
that used the network.

The network card was swapped with a different one, and now FTP zips
files back and forth at 900K/sec; the machine can easily manage X
sessions across the network.

The old card was seen as rl0; the new card is recognised as ed0.

For me personally, the slow network is no longer a problem; it runs
as fast as anyone could expect.  But the drastic difference in speed
is probably due to a driver bug, or something, and this appears to be
an ideal opportunity to track said bug down.  If I know how to do
that, this would be perfect. 8-)

The web page has a "submit a problem report" link, but it seems
reasonable to have as complete a report as possible.  Can someone with
more clues tell me what tests to set up and run that might provide
usable information?

I'm happy to donate my time and CPU cycles to the collective debugging
effort, but how do I diagnose the problem?

"I use not only all the brains I have, but all I can borrow as well."
                                                   -- Woodrow Wilson

Quote:> The old card was seen as rl0; the new card is recognised as ed0.

> For me personally, the slow network is no longer a problem; it runs
> as fast as anyone could expect.  But the drastic difference in speed
> is probably due to a driver bug, or something, and this appears to be
> an ideal opportunity to track said bug down.  If I know how to do
> that, this would be perfect. 8-)

NO, it is NOT a driver problem. It is a _hardware_ problem.

[snip]

Quote:> I'm happy to donate my time and CPU cycles to the collective debugging
> effort, but how do I diagnose the problem?

[0] The comments in there are pretty instructive and even clear to read.
    Read the part in /usr/src/sys/pci/if_rl.c that begins with
    <quote>
      The RealTek 8139 PCI NIC redefines the meaning of 'low end.'
    </quote>

--
  j p d (at) d s b (dot) t u d e l f t (dot) n l .

Quote:>On a machine running 4.5-RELEASE, and the network was just too
>intolerably slow; on a 10baseT LAN, transfer rates averaged 4K/sec.
>This seemed consistent with X, FTP, and just about everything else
>that used the network.
>The network card was swapped with a different one, and now FTP zips
>files back and forth at 900K/sec; the machine can easily manage X
>sessions across the network.
>The old card was seen as rl0; the new card is recognised as ed0.
>For me personally, the slow network is no longer a problem; it runs
>as fast as anyone could expect.  But the drastic difference in speed
>is probably due to a driver bug, or something, and this appears to be
>an ideal opportunity to track said bug down.  If I know how to do
>that, this would be perfect. 8-)
>The web page has a "submit a problem report" link, but it seems
>reasonable to have as complete a report as possible.  Can someone with
>more clues tell me what tests to set up and run that might provide
>usable information?

-------
/*
 * The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is
 * probably the worst PCI ethernet controller ever made, with the possible
 * exception of the FEAST chip made by SMC. The 8139 supports bus-master
 * DMA, but it has a terrible interface that nullifies any performance
 * gains that bus-master DMA usually offers.
 *

-------
That is just a short excerpt.  Further into the driver it documents
the exact problems in the design of this card.  It is
/usr/src/sys/pci/if_rl.c if you want all the*details.

Quote:>I'm happy to donate my time and CPU cycles to the collective debugging
>effort, but how do I diagnose the problem?

Quote:>On a machine running 4.5-RELEASE, and the network was just too
>intolerably slow; on a 10baseT LAN, transfer rates averaged 4K/sec.
>This seemed consistent with X, FTP, and just about everything else
>that used the network.
>The network card was swapped with a different one, and now FTP zips
>files back and forth at 900K/sec; the machine can easily manage X
>sessions across the network.
>The old card was seen as rl0; the new card is recognised as ed0.
>For me personally, the slow network is no longer a problem; it runs
>as fast as anyone could expect.  But the drastic difference in speed
>is probably due to a driver bug, or something, and this appears to be
>an ideal opportunity to track said bug down.  If I know how to do
>that, this would be perfect. 8-)
>The web page has a "submit a problem report" link, but it seems
>reasonable to have as complete a report as possible.  Can someone with
>more clues tell me what tests to set up and run that might provide
>usable information?
>I'm happy to donate my time and CPU cycles to the collective debugging
>effort, but how do I diagnose the problem?

Note: if you get more than 97% packets back, just post final banner -
line (--- bla.bla.bla ping statistics ---) and below.

--
BR, Mishka.

I asked about a problem with a really slow network connection, which
improved drastically when I swapped in a different network card.  I
suggested that some testing might be in order.

"j p d (at) d s b (dot) t u d e l f t (dot) n l" wrote back, directing
me to the source code for the driver.  That was educational, as I've
never wandered into that area of the source tree before.  Bill
Vermillion also quoted some driver source, which was stark in its
assessment of the RealTek 8139 PCI NIC.

Both people took the view that the card wasn't worth the trouble, so I
returned it, explaining to the shop owners that it wasn't up to snuff.
It had come with a computer I bought, and they were happy to exchange
it for another one.  (I've done pretty well with them before; they're
not really BSD-aware, but they're patient about finding hardware I can
work with and swapping parts in and out until we get a match.)

Quote:

>I asked about a problem with a really slow network connection, which
>improved drastically when I swapped in a different network card.  I
>suggested that some testing might be in order.
>"j p d (at) d s b (dot) t u d e l f t (dot) n l" wrote back, directing
>me to the source code for the driver.  That was educational, as I've
>never wandered into that area of the source tree before.  Bill
>Vermillion also quoted some driver source, which was stark in its
>assessment of the RealTek 8139 PCI NIC.
>Both people took the view that the card wasn't worth the trouble, so I
>returned it, explaining to the shop owners that it wasn't up to snuff.
>It had come with a computer I bought, and they were happy to exchange
>it for another one.  (I've done pretty well with them before; they're
>not really BSD-aware, but they're patient about finding hardware I can
>work with and swapping parts in and out until we get a match.)

Bill

--