Re: Router Requirements Inactivity

mahdavi@tesla.psc.edu Sat, 04 June 1994 02:13 UTC

Received: from ietf.nri.reston.va.us by IETF.CNRI.Reston.VA.US id aa14679; 3 Jun 94 22:13 EDT
Received: from CNRI.RESTON.VA.US by IETF.CNRI.Reston.VA.US id aa14675; 3 Jun 94 22:13 EDT
Received: from moe.rice.edu by CNRI.Reston.VA.US id aa24307; 3 Jun 94 22:13 EDT
Received: from tesla.psc.edu by moe.rice.edu (AA19604); Fri, 3 Jun 94 20:52:06 CDT
Received: from localhost by tesla.psc.edu (5.65/Ultrix3.0-C 11/12/92 nydick) id AA14917; Fri, 3 Jun 1994 21:52:00 -0400
Message-Id: <9406040152.AA14917@tesla.psc.edu>
To: Fred Baker <fbaker@acc.com>
Cc: mahdavi@tesla.psc.edu, rreq@rice.edu
Subject: Re: Router Requirements Inactivity
In-Reply-To: Your message of "Fri, 03 Jun 94 13:09:14 -0800." <9406032009.AB17225@fennel.acc.com>
Date: Fri, 03 Jun 1994 21:51:59 -0400
Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: mahdavi@tesla.psc.edu
X-Mts: smtp

I think this summary is pretty close.

>     - Routers SHOULD NOT implement Source Quench
>     - Hosts (including routers) MAY ignore Source Quench
>     - IF Source Quench is implemented
>         - by default don't send it
Actually, I think this is missing in the draft.  I'll add it.

>         - when configured to send it, limit the rate at which you send
>     - All routers DO implement some drop policy (a fact)
>     - Drop policy SHOULD attempt to affect only rogue flows

Hmmm...  I'm not sure that I agree with the implications of the word
"rogue".  Some notion of fairness should apply to the drop policy.
This doesn't mean that the packets you decide to drop are somehow
"bad", though.  This is probably just semantics, however.

>     - Precedence & TOS should be respected in the drop policy
>     - Routers with advanced algorithms must document those algorithms

In fact, it would be useful for all routers to document their drop
algorithm.  Even simple algorithms, such as drop-tail, can be
implemented incorrectly and have ill effects on performance under
congestion.  With a simple algorithm, all that really needs be said is
what the algorithm is, and how long the queues are.  In many routers
the queue-lengths are configurable, and the implication is that the
algorithm is drop tail.  However, some routers have multiple queues at
different places in the architecture, which makes it difficult to
understand how the router is supposed to behave when congested.  (The
basic wish is to be able to distinguish bugs from features :-)

--Jamshid