Yup.

I have similar experience.

I have Netatalk+asun running on a PPC machine, but it used to
run on a IIci, Q610, 7600/132, and now a PC132 w/ 300Mhz upgrade.

I saw similar performance anomalies, and found that for instance,
the transfer speed over the wire changed markedly between the IIci and
the Q610, and then again on the PC132 (PowerCenter 132) before and after
the addition of the upgrade card.  In no case have I seen the netatalk
daemons saturate the CPU on any of these machines.

My conclusion was that there is something in the netatalk processing
that is helped by a faster CPU, but is more on the response/latency
side than on the "processing" side.  If memory serves, I think the
Q610 topped out w/ netatlk at about 500 Kbytes/sec on 10BaseT, and the
IIci topped out about 250K.

When the 7600/132 and PC132/300 both had 10BaseT, each could saturate the
wire, running close to 1 Mbyte/s.  The PC132/300 did a hair better,
though.

The 7600/132 and PC132/300 now both have 100BaseT, and the 7600 runs about
1.5-1.7 Mbytes/sec, while the PC132 runs a little over 2.  This difference
could be due to differences in the drives and/or SCSI on the two machines
(same ethernet card and OS) but result is consistent with the CPU being
a factor.

I'm very curious why I see this behavior, but not (yet) enough to
dig into the netatalk code and find out.

-dgl-

>my 7200/75 (unupgraded) produces up to 800k Bytes/sec over 10 base T 
>from MacOS.
>
>Just to let you know.
>Marty
>
>
>On Wednesday, October 24, 2001, at 05:31 PM, Donald Lee wrote:
>
>> FWIW - I've found that the speed of the processor affects the transfer
>> speed of netatalk beyond what you would expect judging from the
>> CPU stats.  I'm guessing that it's an AFP protocol latency issue.
>>
>> The 7x00 has a slow bus, so the G3 upgrade will help, but it won't bring
>> it up to the standard of the newer machines.
>>
>> -dgl-
>>
>>