HIGMEM, SMP, 2.4.13 and Cerberus

HIGMEM, SMP, 2.4.13 and Cerberus

[Joan Batet]

Hello all,
I've got a little problem with a new computer we have on our company. The 
problem is what I try run Cerberus to test the stability of this computer 
with a 2.4.13 kernel;  at the first test the computer went down in 15 
minutes; the second test went well, the computer responsivity was good and 
it was appears to be stable. At this moment I'm going to test again to see 
if the machine supports it (I hope that ...).
Well, now the problem and the question: what kernel do you think I must 
put on this computer? I need a rock solid system ... I'm testing 2.4.3 and 
2.4.9 too, but 2.4.13 appears to be the best performance kernel ...
Thanks in advance,
	Joan


The computer:
- Dell PowerEdge 2550
- Dual PIII 1Ghz
- 2Gb ECC SDRAM
- 2 x 36Gb RAID1 (AACRAID)

Linux TESTING 2.4.13 #1 SMP Fri Oct 26 12:30:37 CEST 2001 i686 unknown

Gnu C                  2.95.3
Gnu make               3.79.1
binutils               2.10.91.0.4
util-linux             2.11f
mount                  2.11f
modutils               2.4.5
e2fsprogs              1.24
Linux C Library        2.2.2
Dynamic linker (ldd)   2.2.2
Procps                 2.0.7
Net-tools              1.60
Kbd                    1.04
Sh-utils               2.0
Modules Loaded

Re: HIGMEM, SMP, 2.4.13 and Cerberus

[Rik van Riel]

On Mon, 29 Oct 2001, Joan Batet wrote:

> Well, now the problem and the question: what kernel do you think I
> must put on this computer? I need a rock solid system ... I'm testing
> 2.4.3 and 2.4.9 too, but 2.4.13 appears to be the best performance
> kernel ...

2.4.12-ac6 should be able to survive Cerberus.

Rik
-- 
DMCA, SSSCA, W3C?  Who cares?  http://thefreeworld.net/

http://www.surriel.com/		http://distro.conectiva.com/

2.4.0 TCP caches ip route

[Manel Guerrero Zapata]

Hi,

I'm using 2.4.0 (but I thing that this is probably a 2.4.X problem).
If a execute a telnet command to certain address (like 10.0.0.1).
And the routing table says that packets for 10.0.0.1 should be
routed to the device dummy0,
The telnet keeps trying to connect. (till here everything is cool).
And then I change the routing table so now it should
send those packets to the ppp0 (where 10.0.0.1 is), but the
connexion does not get stablished anyway (till I get timeout).

The problem seems to be that the kernel
caches that the device for the connexion should be dummy0.
If then, I cancel the telnet and start it again
now (of course) it stablishes a telnet conexion though the ppp0.

This problem does not occur if I use ping instead of telnet
(probably because ping uses no socket).

Should not be a mechanism that flushes caches when routing
table changes?

If the cached info is attached to the socket structure
probably this flushing thing is not quite feasible, am I wrong?

I know you usually don't expect the device though your tcp
connection goes to change on the fly, but that actually can
happend, and maybe (IMHO) should be supported.

Maybe I'm missing something. (probably ;) )

Regards,

	Manel Guerrero

Re: 2.4.0 TCP caches ip route

[Martin Eriksson]

----- Original Message -----
From: "Manel Guerrero Zapata" <manel.guerrero-zapata@nokia.com>
To: "Kernel Mailing List" <linux-kernel@vger.kernel.org>
Sent: Monday, October 29, 2001 2:03 PM
Subject: 2.4.0 TCP caches ip route


> Hi,
>
> I'm using 2.4.0 (but I thing that this is probably a 2.4.X problem).
> If a execute a telnet command to certain address (like 10.0.0.1).
> And the routing table says that packets for 10.0.0.1 should be
> routed to the device dummy0,
> The telnet keeps trying to connect. (till here everything is cool).
> And then I change the routing table so now it should
> send those packets to the ppp0 (where 10.0.0.1 is), but the
> connexion does not get stablished anyway (till I get timeout).
>
> The problem seems to be that the kernel
> caches that the device for the connexion should be dummy0.
> If then, I cancel the telnet and start it again
> now (of course) it stablishes a telnet conexion though the ppp0.
>
> This problem does not occur if I use ping instead of telnet
> (probably because ping uses no socket).
>
> Should not be a mechanism that flushes caches when routing
> table changes?
>
> If the cached info is attached to the socket structure
> probably this flushing thing is not quite feasible, am I wrong?
>
> I know you usually don't expect the device though your tcp
> connection goes to change on the fly, but that actually can
> happend, and maybe (IMHO) should be supported.
>
> Maybe I'm missing something. (probably ;) )

This sounds like a minor problem. I do think that if you already have an
established connection it would work just as you want it to do, but as you
describe it you don't have an established TCP connection. What I mean is
that you won't exactly loose your connection if you don't even have one at
the first place!

_____________________________________________________
|  Martin Eriksson <nitrax@giron.wox.org>
|  MSc CSE student, department of Computing Science
|  Umeå University, Sweden

Re: 2.4.0 TCP caches ip route

[Manel Guerrero Zapata]

ext David Ford wrote:
> 
> Try "ip route flush cache" or summarized, "ip r f c"
> 
> David
> 
> Manel Guerrero Zapata wrote:
> 
> >The problem seems to be that the kernel
> >caches that the device for the connexion should be dummy0.
> >If then, I cancel the telnet and start it again
> >now (of course) it stablishes a telnet conexion though the ppp0.
> >
> [snipped]


Hi,


Answering to David:

Well, I have not tried that yet. But that is not a real solution.
You are not supposed to be flushing the cache manualy
(Although it might work as kind of workaround)
the kernel should detect that this is a stale cached entry
and delete it.


Answering to Martin:

Sure for most of the people that's not a big deal.
And yes, you are right: if the tcp connection has
been stablished before playing with routing tables
then everything works just fine.

But, IMHO I thing TCP should not make any assumptions about
routing tables (not even during the stablishment of connection).
So, I personaly see this as a kernel bug.

I understand that this is an "optimization" of the kernel
code, but I thing you should have the possibility of
disabling it.


Regards,

        Manel Guerrero

Re: 2.4.0 TCP caches ip route

[Manel Guerrero Zapata]

ext David Ford wrote:
> 
> Try "ip route flush cache" or summarized, "ip r f c"
> 
> David
> 
> Manel Guerrero Zapata wrote:
> 
> >The problem seems to be that the kernel
> >caches that the device for the connexion should be dummy0.
> >If then, I cancel the telnet and start it again
> >now (of course) it stablishes a telnet conexion though the ppp0.
> >
> [snipped]

Hi,

I've tried with:
	echo 0 > /proc/sys/net/ipv4/route/flush 
ip route flush cache is equivalent to: 
	cat /proc/sys/net/ipv4/route/min_delay > /proc/sys/net/ipv4/route/flush

It does not solve anything.
I don't know why.

So not even flushing manually the cache solves
the problem.

Any ideas about why?

	Manel