Re: Bandwidth
From: Gordon Burditt (gordonb.83lk6_at_burditt.org)
Date: 09/30/05
- Next message: Bit Twister: "Re: bourne shell script"
- Previous message: rishi.shah_at_patni.com: "ctrl + c"
- In reply to: rishi.shah_at_patni.com: "Re: Bandwidth"
- Next in thread: Pascal Bourguignon: "Re: Bandwidth"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Fri, 30 Sep 2005 02:07:17 -0000
>I may have created some confusion by my expalnation early on.
>Here are the specifications more clearly
>I am implementing the project in unix/c++ and each node represents a
>process on a separate machine.
>They communicte using TCP sockets
>Each node has a configuration file that has the location co-ordinates.
>When I said 40m distance, I meant the euclidean distance between two
>nodes is 40m.
>
>I have the formula
>Latency = Propagation time + Transmit time + Queing time
>where Propagation time = Distance / Speed of light = m/ (m/s) = seconds
Propagation time is the time from when the first bit is transmitted to
when the first bit is received. It needs to account for *ALL* of
the time-consuming things that happen to the bits on the way from
one node to another. Speed is part of it, but not nearly all of it.
Modulation and de-modulation take time.
If there is a store-and-forward node along the path (e.g. ethernet
switch, or worse, router, or worse yet, mail server with queue),
you might have to add in the transmit time (possibly multiple times).
Processing in the TCP stack takes time.
TCP retransmission due to dropped packets can take a LOT of time,
compared to the other stuff mentioned, especially if you're trying
to calculate a worst-case response time.
>and Transmit time = Packet size/ Bandwidth = bits / (bits/s) = seconds
>I assumed that the Propagation time = 0 since the distance is
>negligible as compared to speed of light.
If the speed of light in whatever medium is involved is only 0.1 c,
propagation time is probably still negligible, especially when
dealing with a transmit time of about 1 millisecond per character.
>Now is my assumption of packets travelling on a TCP socket as fast as
>light incorrect?
In this case, if you assume it's travelling at 0.01 c, it's STILL
negligible. However, passing through a modem may not be negligible.
>Since this project is a simulation I am not aware of the actual
>material between the nodes.
Try throwing in some real numbers and you'll likely find it's
negligible regardless of what material you use within reason.
Let's try an analysis for a different situation: you're sending
the packet by air courier from Dallas to Houston. You have:
(1) Office to airport by car
(2) Getting through security
(3) Boarding time
(4) Flight time
(5) Un-boarding time
(6) Getting out of the airport
(7) Refill gas tank of car
(8) Airport to destination office by car.
(3), (5) and (6) are probably negligible. For a short flight like
Dallas to Houston, (4) is negligible (so the speed of an airplane
DOES NOT MATTER). Post 9/11 (2) probably dominates, (except right
after Rita when (7) may take weeks), so much that driving direct
without the air leg may be faster, and replacing the plane with a
warp-9 space ship won't help.
>Queing time I have assumed to be zero since it is my own simulation.
Even if you assume communication between the two nodes is all that
is going on, queueing may not be negligible. If you're using, say,
100-byte packets at 9600 b/s, it takes 0.1 seconds to transmit a
packet and if you try to transmit a second packet while the first
is still being transmitted, you have to wait. If your protocol
between the nodes is such that this can't happen, (e.g. you won't
even TRY to send a packet until the last one is answered) an
assumption of 0 is OK.
>In such a case is the formula Latency = Packet Size/Bandwidth. Am I
>Correct?
Yes, given assumptions above of what's negligible.
>If I wnat to consider the tranmission time do I have first find the
>material and then the speed of travelling in that material. Am I
>Correct?
No. There's a lot more to transmission time than the speed, although
the difference may be negligible. For instance, I seem to recall
that a propagation time of 0.1 seconds was a pretty good number for
getting data signals through the voice telephone network between
two points in the USA provided no satellites were involved, regardless
of distance (next door vs. coast-to-coast). What's involved is
some of the filters used in the telephone system and in modems,
which dominate speed-of-light issues.
Gordon L. Burditt
- Next message: Bit Twister: "Re: bourne shell script"
- Previous message: rishi.shah_at_patni.com: "ctrl + c"
- In reply to: rishi.shah_at_patni.com: "Re: Bandwidth"
- Next in thread: Pascal Bourguignon: "Re: Bandwidth"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
