At 08:49 PM 7/31/01, Michael L. Williams wrote:
>I'm totally speculating here........ Please let me wrong if this doesn't
>jive....... it seems to me that the number of packets in the queue
>(outgoing) wouldn't be anymore for 50% or 99% until there are more packets
>attempting to go over the wire than is allowed.... (i.e. there won't be any
>use for the queue except in passing as the packets are sent out as fast as
>they are coming to the interface to be sent out).......
Consider a restaurant. As usage of the restaurant increases, the delay to
get your food increases also. This happens even if the restaurant isn't
full. I'm sure we have all experienced this! ;-) How bad it gets depends on
the arrival rate of patrons and how long they stay and demand service. The
arrival rate needs to be considered not just in terms of an average rate,
but also in terms of how many patrons arrive at once.
There's a bunch of different ways of figuring out queue depth, M/M/1,
M/D/1, etc. See David Spohn's book, "Data Network Design" for a good
description. The example I used was a M/M/1 example, I think. Utilization
is dimensionless in this method and is directly related to the average
number of arriving bursts (packets, frames, or cells) per second. It is
also related to the average number of seconds per burst. Or something like
that. Bottom line: queuing delay is complex and it's time for dinner.
Here's something else to chew on, while using a food analogy:
As an example, it takes 5 ms to transmit a 1024-byte packet on a 1.544-Mbps
T1 link. That's a given. It has nothing to do with statistics or averages
or anything fuzzy. There's serialization delay on all non-parallel ports,
although on high-speed links it is minimal.
There's a good chance a packet will get queued during that 5 ms, especially
if there's a bunch of Ethernet segments feeding into this T1 link. It
depends on traffic bursts, length of bursts, etc., as Jenny pointed out.
See you tomorrow. I'm signing off! For dinner, I'm having a burst of beets.
Their arrival rate in my garden has gotten out of control and something has
to be done.
Priscilla
>I read Jenny's post and I agree with her that utilization is = packets/some
>amount of time...... and that things tend to get bursty, etc..... so if
>depends on if you're looking at a 1 sec avg. or a 1 minute avg. utilization
>as to whether a ping at any given moment would actually go through as fast
>as any other time.
>
>Mike W.
>
>"Priscilla Oppenheimer" wrote in message
>[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> > The number of packets in a queue on a packet-switching device increases
> > exponentially as utilization on the output port increases.
> >
> > Queue depth = utilization/(1 - utilization)
> >
> > So, do the math. If utilization is 90%, there will be more packets in the
> > queue than if utilization if 50%.
> >
> > That's how I learned it, but it's probably more complex than this....
> >
> > Comments, anyone else?
> >
> > Thanks
> >
> > Priscilla
> >
> > At 10:41 AM 7/30/01, anil.philip wrote:
> >
> > >Dear Priscilla,
> > >
> > >I have a small doubt. I think you are the best person to ask about. May
>be
> > >you think this as a stupid question. but this is making me crazy.
> > >
> > >If there is a serial link b/w two sites, at what point of % utilisation
> > >the response start degrading???
> > >
> > >Ideally if I have a T1 link, i shud get the same ping response time
> > >till the load on that link is 100%???
> > >
> > >When I say a T1, it is 1.55 M packets /s. So if the link is utilised for
> > >99% (say 1.50Mb..) still I have
> > >0.05 Mb left on that link and I shud be able to get a ping (32byte)
> > >response time, equivalent to 0% utilisation. Why the response time start
> > >degrading at some point of % utilisation??
> > >Let us assume there is no packet drop, memmory prob, enough buffer space
> > >etc....
> > >
> > >Request to do a REPLY ALL this email.
> > >
> > >
> > >Regds,
> > >Anil Philip
> > >AT&T Solutions.
> > >anil.philip
> > >e-mail: [EMAIL PROTECTED]
> >
> >
> > ________________________
> >
> > Priscilla Oppenheimer
> > http://www.priscilla.com
________________________
Priscilla Oppenheimer
http://www.priscilla.com
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=14514&t=14233
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
