Re: We want to prevent packet lost? But TCP needs packet lost.

From: Gary Duncanson (gary.duncanson@googlemail.com)
Date: Sun Jul 22 2007 - 09:11:02 ART

You still need QoS. If buffers fill up packets get dropped. Tail drop.

WRED is one approach to get over this, discarding packets packets as a queue
begins to fill rather than waiting for the impact of tail drop and multiple
packet loss. WRED attempts this to prevent TCP connections from really
slowing down.

On packet loss, TCP senders slow down their rate of sending data. It has
it's dynamic windowing mechanism and such like to help, and the ACK
sequencing and retransmit thing.

Even so, if it takes 'too long' for packets to get there due to congestion
on links your application may be too slow to be useful even with the native
TCP features. So you still need QoS to prioritise the time sensitive stuff.
Then there is voice and the jitter problems.

Then again, if the QoS is poorly implemented you may drop the important
stuff by design!

I'm still waiting for Wendall Odom's QoS book to arrive. I expect many happy
hours ahead labbing QoS features.

Gary

----- Original Message -----
From: <johngibson1541@yahoo.com>
To: <ccielab@groupstudy.com>
Sent: Sunday, July 22, 2007 4:24 AM
Subject: Re: We want to prevent packet lost? But TCP needs packet lost.

> Even smarter these people. The TCP protocol is designed to
> have packet lost. And they can just come up with more and more
> sophisticated queues every month to "prevent packet lost".
>
> It is designed to have packet lost, and they come to their
> work every day to "prevent packet lost". How smart is that.
>
> They will never need to create any thing and they have
> a job guaranteed for life.
>
> This is like the lawyers working with law makers to come up
> with more sophisticated forms to do the same thing every
> month so that citizens need lawyers.
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html

This archive was generated by hypermail 2.1.4 : Sat Aug 18 2007 - 08:17:41 ART