On Sat, Oct 07, 2006 at 03:35:48PM +0200, Thomas Feddersen wrote:
> My server synchronizes data with another server in a different location  
> over the internet (dsl 128kbit/sec) at regular intervals. I use ssh to  
> tunnel through the intenet. The software used is rsync and / or unison.  
> The server is connected to the internet over a hardware router (ASUS  
> WL500g).
> 
> Whenever the system is actively transferring data, is responds very  
> sluggishly to requests form the LAN: starting Midnight Commander on the  
> server takes about 30 sec, opening an Explorer window on a samba client  
> (WinXP or OSX) takes something between 15 sec and 60 sec. Normally either  
> of these commands are finished in just 2 sec.
> 
> The system seems not to be loaded: 'top' shows load-averages is well below  
> 1 and an idle of over 90%. The process that does the synchronizatiuon  
> utilizes about 1% CPU. 'systat vmstat' shows hardly any swapping or  
> paging. 'systat iostat' shows no disk-activity beyond the data-stream to  
> and from the interet.
> 
> I think NetBSD is supposed to be able to handle many network-connections  
> simultaneously. Is this at all a network problem or is maybe ssh clogging  
> something in the kernel? The server does have a software-RAID5 array  
> (Promise IDE controller in PCIbus), but the synchronization pertains only  
> to the drive where the NetBSD system resides (onboard PIIXIDE).
> 
> Does anybody have an idea, how to remedy this problem?

Can you properly ping the box ? open a ssh connection ? resolve hostnames ?

To me it looks like a DNS issue: while you're running the transfers over
the DSL line it's overloaded, and DNS queries can take a long time.
The solution would be to run a caching nameserver locally.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--