Talk:Throughput

From Wikipedia, the free encyclopedia

"Bus" network topology This article is part of WikiProject Computer networking, an attempt to build a comprehensive and detailed guide to Computer networking on Wikipedia. If you would like to participate, you can edit the article attached to this page, or visit the project page, where you can join the project and/or contribute to the discussion.
Computer networking Portal
Start This article has been rated as start-Class on the assessment scale.
Computer networking To-do:
WP:TEL This article is within the scope of WikiProject Telecommunications, an attempt to build a comprehensive and detailed guide to telecommunications on Wikipedia. If you would like to participate, you can edit the article attached to this page, or visit the project page, where you can join the project as a "full time member" and/or contribute to the discussion.

Contents

[edit] Latency

(heading added later for clarity --Alf 11:15, 11 January 2007 (UTC))

The reference "It's the latency, stupid" asserts speed is distance divided by time, and bits are not distance, so bits/second is not speed. The speed of a runner may be distance divided by time, but the speed of a worker is not distance divided by time.

He points out the difference between the throughput and the speed of an oil tanker. It may be slow but still have a large capacity. I think that a more apt comparison would be to a pipeline, where the notion of 'serialization' can be applied.

[edit] Latency 2

(heading added later for clarity --Alf 11:15, 11 January 2007 (UTC))

I'm sure that latency is an important problem in the flow of bits and bytes through a computer and I would no doubt enjoy an entry devoted to the topic. The entry's title is Throughput, though, and in the Theory of Constraints the word is defined as: the money (not the output) an organization receives in exchange for the goods or services its customers (clients, students, patients, ...) receive. The signal (money) is received (or not) at the best time and place, and using throughput instead of output (goods and services produced but not necessarily sold) prevents the latency problem addressed in the entry. See, for example, Eliyahu M. Goldratt and the List of Theory of Constraints topics.

[edit] Bandwidth

(heading added later for clarity --Alf 11:15, 11 January 2007 (UTC))

I would like to see a closer comparison between throughput and bandwidth. The differences, similarities and the common usages between the two.

Throughput is the actual speed data will transfer at from one point on the network to another. The actual speed, throughput, of a network is often quite less than it's Bandwidth, potential speed.

Bandwidth is more often quoted than Throughput since it is easier to calculate and not subject to changes in many variables. The hardware used to communicate, heavy/light traffic, etc…can effect throughput.

[edit] Too general a title?

throughput is used in a lot of different contexts to describe the rate at which a process works -- for example, throughput of an assembly line, throughput of a polling place, throughput of a communications channel (slightly more general than the discussion of internet connections here), and throughput of a CPU (quite relevant to the latency/throughput dichotomy raised in the article). There could/should be some discussion of queueing theory and the importance of throughput reserve to reduce queues; of traffic on roads (maximum throughput of a freeway occurs when the traffic is so heavy it drops to about 30 MPH); and of things like the throughput/power relation in engines. zowie 18:50, 21 November 2005 (UTC)

I agree with you that the title is too general. For instance, the Latency page is a disambig page and the Latency related to this article is Latency (engineering). The article titled "Throughput" should be a disambig page like Latency and this page should be renamed to "Throughput (communications)" or some other descriptor. -Etienne 08:43, 10 August 2006 (UTC)
Any objections on reaneming this article to something like Throughput (communications)? Any other name suggestions? --Alf 11:20, 11 January 2007 (UTC)


Shouldn't there also be under latency a reference to Round Trip Time (RTT)?

[edit] OIl and Natural Gas

Throughput is a term used in the pipeline industry; in case someone knows how to put it in the article. --McTrixie 00:35, 17 October 2006 (UTC)

[edit] Rename

Refering to Talk:Throughput#Too general a title? this article should be renamed to something like Throughput (communications) and a disambiguation page written for Throughput. Please offer your suggestions for names. --Tunheim 08:57, 7 February 2007 (UTC)

Okay. Don't forget to redirect all pages pointing here. There are about 150 pages. Mange01 12:59, 20 August 2007 (UTC)


[edit] Context Expansion Beyond Pure Digital Data?

In an information theory context Throughput refers to messages, and has massively varied units, such as chips/s (in the case of gold-coded wireless networks) or frames/sec, in the case of some analog networks. Would it make sense to edit the article to recast it in more general terms and then treat the different units of throughput as subsections? (I'm a new user and don't want to disrupt work in progress) HatlessAtless (talk) 20:54, 23 April 2008 (UTC)

This is interesting. I have never seen chips/s used as a throughput measure - only as a measure of the chip rate, which is a baud rate that typically is constant, since the bandwidth in Hertz is constant. Throughput for me is affected by traffic load, and mechanisms such as packets queues, retransmissions, and multi-access protocols. The spreading factor (chips/bit) may be adapted to the channel (the distance between the nodes), and sometimes to the bit rate or Quality of Service level, which may be negotiated based on the traffic load. But I don't see why it would be be interesting to study the throughput in chips/s.
Regarding your question: If you intend to remove something, I suggest that you discuss it here first. Otherwise feel free to reorganize the article. If you are unsure, you may also suggest a new disposition here first. Mange01 (talk) 10:08, 24 April 2008 (UTC)
Thanks for the advice. I will indeed reorganize the article. Comments on my rework style will be greatly appreciated, however. HatlessAtless (talk) 21:49, 24 April 2008 (UTC)
Thnx for your improvements to the article! Since you ask for comments, I answer, by don't take it as criticism, you have done a great job.
Is it possible either to change the order of the text, in view to put practical stuff and key definitions that is interesting to a large audience in the beginning, or to rephrase the theoretical definitions to make them more comprehensible to the average reader?
What is the practical area of application of these maximum TP definitions? To my understanding, max TP may be one of the following:
1. Something you measure in a real-world network using various bandwidth test services. This is interesting to a large audience, and is discussed in the Measuring network throughput article. It would be good if it was mentioned early in this article. This measure is affected by traffic load from other users. Typically it is measured below or above OSI layer three, i.e. exclusive of Ethernet overhead/PPP overhead, and sometimes exclusive of IP overhead. I don't know if this is what you call Peak Measured Throughput or Maximum Sustained Throughput.
2. A capacity measure that you can calculate on paper by means of formulas (which you call Maximum Theoretical Throughput, and is called "Maximum throughput over analog channels" in the old version of the article).
3. An idealized measure that scientists evaluate in computer simulations or measure in experiment networks, for example in view to compare various algorithms. I guess this is what you have in mind in "Maximum achievable throughput" section, or? This kind of measure highly depends on the traffic model, which sometimes is simplified and idealized, and sometimes aim at reflecting "real" traffic situation. For example, you may assume "real" flow control and congestion avoidance (TCP traffic), simplified flow control model (one packet is generated for every packet that is delivered over a certain connection), no flow control (UDP traffic), Erlang traffic (with guaranteed quality of service), Poisson packet source (exponentially distributed time between packets), bursty self-similar traffic model, etc. It is also affected by what router model and traffic shaping algorithms we assume. I have seen several alternative definitions:
    • Asymptotic maximum throughput - the load when the delay goes to infinity, or the load when the system becomes unstable, and less packets are delivered to the destination than arrive from the source. Very hard to measure, since it requires infinite packet queues.
    • Maximum throughput for a certain drop rate, for example 5%, where packets older than a certain time-to-live are dropped.
    • Maximum throughput assuming the above simplified flow control model. Very simple to simulate.

Mange01 (talk) 21:39, 27 April 2008 (UTC)

Thanks for the feedback!
The intent I had when breaking down the different max throughput was to make the point that all "max throughput' numbers are not created equal and give a broad flavor of the different possible meanings. If you think that my sections should be reorganized, feel free to shuffle them as makes the most sense. HatlessAtless (talk) 16:52, 28 April 2008 (UTC)

[edit] Unsourced Tag During Rework?

It will take me several days to organize and source this article fully. Should I leave the unsourced tag alone until I complete the process? Also, is it appropriate to flag my own posts for copy editing until a more experienced editor can look at my work? HatlessAtless (talk) 21:49, 24 April 2008 (UTC)