where is k_compat.h?

[Michael Gibson]

A couple of other questions need to be answered before we can "go public"
with this problem:

1. If we back up to a 2.0.x series kernel, does the problem disappear?

2. Do other NICs suffer from the same full-duplex problem as 3COM NICs?

Here's a brief description of what I've experienced relating to this issue.
My Linux box is a P-200 64Mb running Slackware 7.0 (kernel 2.2.13). I have
an ADSL modem going into one vortex card and the switch going into another.
I'm using IP masquerading, so my PCs on the LAN can all use the Net.  I
installed a LinkSys 8 port 10/100 workgroup switch a few months ago.

Here's what I'm 100% sure of:
()Periodically, I notice that throughput from a PC to my Linux box has
dropped severely.  To fix it, I reset power to the switch.
()The full duplex LED on the switch comes ON when my Linux box is first
powered on (before Linux boots), but then after booting performance is
pretty bad.  So I unplug and replug power to my switch and guess what?  The
full duplex LED stays OFF!
()When throughput to my Linux box drops severely, performance is fine
between Windows machines using the same 3com cards (in full duplex as well)
on the switch.
()I have another Linux box with kernel series 2.0.x setup for a customer and
it is connected to a similar model 24-port LinkSys 10/100 switch.  There are
never any performances issues there.  It also uses IP masquerade to the
Internet, but its a dialup connection, rather than ADSL-to-Ethernet.  So
this would seem to answer question #1 from above.




> -----Original Message-----
> From: owner-linux-vortex-bug@beowulf.org
> [mailto:owner-linux-vortex-bug@beowulf.org]On Behalf Of James Ralston
> Sent: Wednesday, February 02, 2000 7:42 PM
> To: pedwardson@mrcnh.com
> Cc: linux-vortex-bug@beowulf.gsfc.nasa.gov
> Subject: Re: where is k_compat.h?
>
> Moderately heavy traffic to where?  To a machine on the same switch,
> elsewhere on your LAN, or elsewhere in the world?
>
> What types of connections slow to a creep and then stall?  TCP/IP
> connections, all IP connections, or all connections of any type?
>
> I ask because I have a tentative theory that the TCP/IP stack in
> recent Linux kernels has a pathological case where throughput becomes
> abysmal, and one of the situations where this seems to occur is with
> two machines connected to the same switch talking to each other as
> fast as they can in full duplex.
>
> I'm still testing, so I'm reluctant to make any concrete statements,
> but you might want to try one or more of the following things:
>
>     Try turning off the RFC1323/RFC2018 features of the Linux 2.2.*
>     kernel, by doing the following:
>
>         # cd /proc/sys/net/ipv4
>         # echo 0 >tcp_sack
>         # echo 0 >tcp_timestamps
>         # echo 0 >tcp_window_scaling
>
>     I suspect doing so will prevent the connection from stalling
>     entirely, but the throughput will probably still slow to a crawl.
>
>     Try forcing both the switch and the NIC to half duplex mode.  In
>     my limited testing, I've found that if I force the NIC of the
>     machine that is *sending* the data to half duplex, the performance
>     problems mostly vanish.
>
> I'd be very curious to know what results you see...
>
> Regards,
> James
>
> -------------------------------------------------------------------
> To unsubscribe send a message body containing "unsubscribe"
> to linux-vortex-bug-request@beowulf.org
>

-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-vortex-bug-request@beowulf.org