This is a multi-part message in MIME format.
--------------090304020503000600090900
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Johnny Billquist wrote:
> Hans Rosenfeld wrote:
> 
>> This smells like the problem I had with my 4k90, all packets > 4k would
>> be corrupted, see 
>> http://mail-index.netbsd.org/port-vax/2006/03/08/0009.html
>>
>> I have still no clue where to look for this.
> 
> 
> No, I would say this is definitely not the same thing.
> Note that here we're talking about a case where I have limited NFS 
> transfer size to 2K, and *then* I have a problem with every other 4K 
> sized piece.
> 
> If I run a default generic kernel, where NFS sizes haven't been 
> restricted, all works just fine.
> For that kernel, it looks like NFS is doing 8K writes.
> 
> Whatever problem you had, I think that has been resolved, Hans.

Interesting thing, though. The driver have some "problem". Note the 
response times...
And Junk is a plain vanilla i386 running NetBSD, while Gnat is a VS4K90 
running NetBSD -current.
And the VAX is otherwise totally idle at the moment.

Junk:/usr/src# ping -s 5000 gnat
PING gnat.BQTnet.SE (10.0.0.17): 5000 data bytes
5008 bytes from 10.0.0.17: icmp_seq=0 ttl=255 time=300.476 ms
5008 bytes from 10.0.0.17: icmp_seq=1 ttl=255 time=1018.466 ms
5008 bytes from 10.0.0.17: icmp_seq=2 ttl=255 time=12.687 ms
5008 bytes from 10.0.0.17: icmp_seq=3 ttl=255 time=151.606 ms
5008 bytes from 10.0.0.17: icmp_seq=4 ttl=255 time=124.407 ms
5008 bytes from 10.0.0.17: icmp_seq=5 ttl=255 time=1018.575 ms
5008 bytes from 10.0.0.17: icmp_seq=6 ttl=255 time=12.746 ms
5008 bytes from 10.0.0.17: icmp_seq=7 ttl=255 time=1018.647 ms
5008 bytes from 10.0.0.17: icmp_seq=8 ttl=255 time=13.461 ms
5008 bytes from 10.0.0.17: icmp_seq=9 ttl=255 time=1018.634 ms
5008 bytes from 10.0.0.17: icmp_seq=10 ttl=255 time=12.833 ms
5008 bytes from 10.0.0.17: icmp_seq=11 ttl=255 time=1018.575 ms
5008 bytes from 10.0.0.17: icmp_seq=12 ttl=255 time=12.760 ms
5008 bytes from 10.0.0.17: icmp_seq=13 ttl=255 time=212.940 ms
5008 bytes from 10.0.0.17: icmp_seq=14 ttl=255 time=1018.737 ms
5008 bytes from 10.0.0.17: icmp_seq=15 ttl=255 time=12.864 ms
5008 bytes from 10.0.0.17: icmp_seq=16 ttl=255 time=1018.492 ms
5008 bytes from 10.0.0.17: icmp_seq=17 ttl=255 time=12.696 ms
^C5008 bytes from 10.0.0.17: icmp_seq=18 ttl=255 time=251.002 ms

----gnat.BQTnet.SE PING Statistics----
19 packets transmitted, 19 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 12.687/434.769/1018.737/466.040 ms
Junk:/usr/src#


Hmm. Checking a bit more, the timing problems appear when the packet 
gets bigger than 2004 bytes. Hmm, that's 44 bytes from 2K. Could that be 
the extra overhead of IP, ICMP and PING perhaps?

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt@softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol

--------------090304020503000600090900
Content-Type: text/x-vcard; charset=utf-8;
 name="bqt.vcf"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
 filename="bqt.vcf"

begin:vcard
fn:Johnny Billquist
n:Billquist;Johnny
email;internet:bqt@softjar.se
tel;cell:+46 70-8641486
x-mozilla-html:FALSE
version:2.1
end:vcard


--------------090304020503000600090900--