Ascend Archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: (ASCEND) Pool handling strategies (was: IP Addresses needed for 4048)
On Wed, Apr 01, 1998 at 03:58:43PM -0500, Phillip Vandry wrote:
>
> One last question, and then we'll call this off topic :-)
>
> But then are are TCP keepalives for? Aren't they used to detect failed
> connections and close them?
As to my knowledge there is no such thing as a TCP keepalive. Application
layer protocols based on TCP may decide to use keepalives, but thats a
problem of these protocols. Generally keepalives are BAD (broken as
designed) anyway.
> In my experience this works UNLESS one of the sides tries to send data
> to the other side. If that happens then it will usually time out and close
> itself within a few hours. Maybe that's newer stuff that wasn't around in
> the BBN day?
If the side that is still alife does not have any application layer
timeout it will send for a really long time. I have seen dangling TCPs
trying to get a segment acked for _weeks_ - we needed to filter them in
order to not make the customer poor and the telco a golden nose. They
continued despite the fact that our Max bombed every segment back with
a destination host unreachable message. It finally stopped, probably the
other side rebooted their server. Was some HPUX box IIRC. The reason for
the dangle was rather esoteric: The peer at our customer was rebooted and
did the correct thing on these segments: it replied with a RST. However
Sun seemed to code the "inplace packet handling" wrong in their kernel,
the RST reply went out with the TTL of the incoming TCP segment, not with
the TTL the Sun usually uses to originate packets. Now this packet was
sent on the HPUX side with a very low TTL (32) and due to the roundtrip
beeing more than 32 hops the RST never reached the other side but starved
on the half way to it. Shit happens ;)
--
Kanther-Line: PGP SSH IDEA MD5 GOST RIPE-MD160 3DES RSA FEAL32 RC4
+-o-+--------------------------------------------------------+-o-+
| o | \\\- Brain Inside -/// | o |
| o | ^^^^^^^^^^^^^^ | o |
| o | Andre' Beck (ABPSoft) beck@ibh-dd.de XLink PoP Dresden | o |
+-o-+--------------------------------------------------------+-o-+
++ Ascend Users Mailing List ++
To unsubscribe: send unsubscribe to ascend-users-request@bungi.com
To get FAQ'd: <http://www.nealis.net/ascend/faq>
Follow-Ups:
References: