Okay, it seems to be fixed now.  I'm running a -current kernel with
ragge's "FAST_MBSEARCH" option.  Maybe it's it, or it's Manuel's fix
to the Promise card that fixed it, don't know.

Client -> Server: ~70Mbit/s without stalls, I'm happy with that value.
Server -> Client: ~100Mbit/s without stalls.

A strange thing tho, increasing net.inet.tcp.*space above default values
makes the transfers unstable, and lot of stalls which kick the transfer
down to ~25Mbit/s, then rising again... Can't explain that, but, my
problem is solved :)


Thanks you who have helped me debug.


Regards,
Mattias.

Manuel Bouyer wrote:
> On Tue, Jul 01, 2003 at 11:49:46PM +0200, Mattias Karlsson wrote:
> 
>>Hi,
>>
>>I've been searching for the probably cause for a while now, and I
>>now give up.  This is my setup:
>>
>>SERVER
>>	Intel P3 500MHz
>>	224MB RAM
>>	4GB IDE disk for / and swap
>>	3Com 3c905B-TX 10/100 Ethernet (rev. 0x30)
>>	Promise Ultra100TX2/ATA Bus Master IDE Accelerator (rev. 0x02)
> 
> 
> BTW, I just commited a fix for performances on Ultra/100, but it shouldn't
> hurt you that way.
> 
> 
>>		wd1 at pciide1 channel 0 drive 0: <IC35L080AVVA07-0>
>>		wd1(pciide1:0:0): using PIO mode 4, Ultra-DMA mode 5
>>		(Ultra/100) (using DMA data transfers)
>>
>>		wd2 at pciide1 channel 0 drive 1: <IC35L100AVVA07-0>
>>		wd2(pciide1:0:1): using PIO mode 4, Ultra-DMA mode 5
>>		(Ultra/100) (using DMA data transfers)
>>
>>		wd3 at pciide1 channel 1 drive 0: <Maxtor 6Y200P0>
>>		wd3(pciide1:1:0): using PIO mode 4, Ultra-DMA mode 5
>>		(Ultra/100) (using DMA data transfers)
>>	OS: NetBSD 1.6.1
> 
> 
> 
> 
>>  Bits/second sent :     23688624  (23.69 Mbit/s)
>>
>>^-- A *lot* of stalls and no stable transfer rate.
> 
> 
> Maybe a tcpdump from the server would help us understand what's going on.
> 

-- 
Mattias Karlsson
mattias.karlsson@sergei.cc
SysAdm - http://www.sergei.cc/