Hi all,

I have a problem with my AIX 4.3.3 server. The TCP flow control algorythm
seems to operate one way only : packet sizes are decreased as low as 20 data
bytes per packet but never goes up again once this is done. This has
drawbacks with some IP clients.

Does someone know how to control this ?

Regards,
Sylvain.

"SCo" <sylvain@coutant.info.nospam> schrieb im Newsbeitrag
news:400fd1ad$0$7139$626a54ce@news.free.fr...
> Hi all,
>
> I have a problem with my AIX 4.3.3 server. The TCP flow control algorythm
> seems to operate one way only : packet sizes are decreased as low as 20
data
> bytes per packet but never goes up again once this is done. This has
> drawbacks with some IP clients.
>
> Does someone know how to control this ?
>
> Regards,
> Sylvain.
>
Hallo Sylvain,

IIRC path MTU values that where decreased are checked every 10 minutes,
while increases are checked every 30 minutes. Thus I doubt that it is just
your AIX server's fault that the packets really stay small after they where
decreased. Check your pmut_rediscover_interval and of course if the route in
question does not support RFC 1191 it migt also cause probs.
tcp_pmtu_discover is default enabled from 4.3.2 on.

HTH,
Andreas

> IIRC path MTU values that where decreased are checked every 10 minutes,
> while increases are checked every 30 minutes. Thus I doubt that it is just
> your AIX server's fault that the packets really stay small after they
where
> decreased.

But they are not decreased because of PMTU but only because network became
slow during a few milliseconds. Nothing to do with PMTU. In case the network
become fast again, the packet size should increase back the same way it was
decreased when the network became slow. Obviously, this is not the case.

Regards,
Sylvain.

"SCo" <sylvain@coutant.info.nospam> schrieb im Newsbeitrag
news:4010eefc$0$22300$626a54ce@news.free.fr...
> > IIRC path MTU values that where decreased are checked every 10 minutes,
> > while increases are checked every 30 minutes. Thus I doubt that it is
just
> > your AIX server's fault that the packets really stay small after they
> where
> > decreased.
>
> But they are not decreased because of PMTU but only because network became
> slow during a few milliseconds. Nothing to do with PMTU. In case the
network
> become fast again, the packet size should increase back the same way it
was
> decreased when the network became slow. Obviously, this is not the case.
>
> Regards,
> Sylvain.
>

I already understood that your packet size stays small and I hinted to some
areas for further investigation. It does not help when you just pretend that
the packets stay small without fostering this thread with further details.
That packets do not increase again is still no prove that it is your
server's problem. Therefore I would suggest you to post the output of:
# no -a |grep pmtu
# lslpp -l bos.net.tcp.*
# lsattr -El ent0 (use your adapter in question)
# lsattr -El en0 (use your adapter in question)
I would also be interested in what type of network you use that becomes
intermittent slow and then fast again. Does it become slow again within the
30 minute pmtu_rediscover_interval? Did you change the interval to 1
minute? Did that change anything?

Regards,
Andreas

SCo <sylvain@coutant.info.nospam> wrote:
> But they are not decreased because of PMTU but only because network
> became slow during a few milliseconds. Nothing to do with PMTU. In
> case the network become fast again, the packet size should increase
> back the same way it was decreased when the network became
> slow. Obviously, this is not the case.

AIX TCP alters segement size during congestion?!? Are you sure you
aren't really taking about the size of the congestion window and not
the packet size?

rick jones
--
oxymoron n, commuter in a gas-guzzling luxury SUV with an American flag
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...

> AIX TCP alters segement size during congestion?!?

Looks like it does. I have 1 GB of ethereal capture to see it :-)

> Are you sure you
> aren't really taking about the size of the congestion window and not
> the packet size?

I am talking about packet size.

Sylvain.

> I already understood that your packet size stays small and I hinted to
some
> areas for further investigation. It does not help when you just pretend
that
> the packets stay small without fostering this thread with further details.
> That packets do not increase again is still no prove that it is your
> server's problem. Therefore I would suggest you to post the output of:
> # no -a |grep pmtu
> # lslpp -l bos.net.tcp.*
> # lsattr -El ent0 (use your adapter in question)
> # lsattr -El en0 (use your adapter in question)
> I would also be interested in what type of network you use that becomes
> intermittent slow and then fast again. Does it become slow again within
the
> 30 minute pmtu_rediscover_interval? Did you change the interval to 1
> minute? Did that change anything?

I will send what you want asap.
I have no prove it is my server problem, computers connecting to the server
are Windows (several versions) FTP clients. They are not often connected
more than a few minutes, so 30 minutes is surely a very long delay for them.

Thanks for your help.
Sylvain.

SCo <sylvain@coutant.info.nospam> wrote:
>> AIX TCP alters segement size during congestion?!?
> Looks like it does. I have 1 GB of ethereal capture to see it :-)

I take it the packet size decreases correlate to TCP retransmissions?
I'd heard of the prospect of "guessing" PTMU's based on
retransmissions, but didn't realize that anyone actually implemented
it.

rick jones
--
Wisdom Teeth are impacted, people are affected by the effects of events.
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...