Subject: Re: Are tlp[01] weak for big traffic?
To: None <port-cobalt@netbsd.org>
From: NAKAJI Hiroyuki <nakaji@jp.freebsd.org>
List: port-cobalt
Date: 12/17/2006 11:27:27
The problem still exists...

>>>>> In <8664cjswai.fsf@ra333.heimat.gr.jp> 
>>>>>	NAKAJI Hiroyuki <nakaji@jp.freebsd.org> wrote:
> >>>>> In <87d56u22id.fsf@roddy.4407.kankyo-u.ac.jp> 
> >>>>>	NAKAJI Hiroyuki <nakaji@jp.freebsd.org> wrote:

> >    -----------+----------- main stream (connected to the Internet)
> >               |tlp0        OCN registered /29 network
> >             +-+-+  \
> >             |RaQ|  NAT
> >             +-+-+  /
> >               |tlp1
> >      ---+-----+----------- private subnet (testing network, 192.168.1.0/24)
> >       +-+-+
> >       |P C|
> >       +-+-+

> > > > My assumption is following. What is the possibility?
> > > >
> > > > 1. RaQ 2 is too slow to handle large traffic.
> > > > 2. The HUB I'm using on private subnet is too weak.
> > > > 3. tlp(4) has some problem.
> > > > 4. other
> > > >
> > > > If 3 is the true reason, what shoud I do to debug or improve it? Do I have
> > > > to show more information about my RaQ2?

> The most probability seems #4. I changed the NIC's driver to that
> included in Windows itself and I also changed the settings to default,
> and the performance seems stable.

> Good news. NetBSD/cobalt is not bad. Thanks.

I have to say "bad news", sorry. NetBSD/cobalt may have a problem
around tlp(4).

I tested again.

Settings:
1. network construction like above, IPv6 also available
2. observe the throughput of tlp[01] with SNMP and MRTG

Try:
1. Download from Win some large files, i.e, Solaris Express DVD
   images.

Result:
1. download speed gets 400kB/s in a few seconds, keeps it about an
   hour, and decreases rapidly to 30-40kB/s
2. After all download never continues though it is online
3. At the same time, some network application loses their all
   connections, for example, smbd/nmbd is working but no SMB
   connection, nfs mounted partition gets unavailable and so on
4. But tlp[01] are all UP and IPv6 connection can be available

The only solution I could do is to restart Cache RaQ 2 only, and
because of nfs failure, I have to break into db and type "sync".

And, "ifconfig tlp0 down; sleep 10; ifconfig tlp0 up" does not help.

Next step:
I will try an off the shelf NAT device as Robert mentioned.
-- 
NAKAJI Hiroyuki