[ltp] cpu load on PCMCIA usage

[Maik Musall]

Hello all,

I've been using Linux with kernels from 2.4.4 to 2.4.19 on my 600X for years
now, and everything's working, but all those years there's one thing that
still annoys me.

I have five different pcmcia network devices in use.
1. Lucent Wireless LAN, 11 MBit
2. Linksys 10/100 MBit+Modem
3. D-Link 10 MBit
4. 3com 3c589, 10 MBit
5. Silicom 10 MBit

All do work basically, but when transferring at about 1 MByte/s (10 MBit),
the cpu load is always at almost 100% system time. As a consequence,
I did not manage to transfer more than 1.4 MByte/s over the 100MBit
Linksys card instead of 7-9 MByte/s. The cpu is the bottleneck.

When I transfer a large file to my K6-233 with Intel EtherExpressPro-100,
that machine idles around at 5% load, while my PIII-500 600X Thinkpad
sticks at 100% cpu, limiting the transfer rate.

Protocol does not matter, occurs with ftp as with rcp or nfs or even
a simple netcat from /dev/zero on one to /dev/null on the other side.
Also, direction of transfer does not matter.

Is that normal? Are there parameters for the i82365 that I don't know
of that may improve this? Or any other hint where to look?

Here's part of my config.opts:
------- snip
include port 0x100-0x4ff, port 0x800-0x8ff, port 0xc00-0xcff
include memory 0xc0000-0xfffff
include memory 0xa0000000-0xa0ffffff, memory 0x60000000-0x60ffffff
include port 0xa00-0xaff
exclude irq 4
exclude irq 3
exclude irq 7
------- snap

Thanks and regards
-- 
Maik Musall, maik@musall.de, http://www.musall.de
PGP fingerprint: 99 F6 3D F5 BA 42 50 94  01 36 FA A9 27 9D 3C 39