Talk:Traffic shaping/Archive 1

BitTorrent
Perhaps this section should be changed to ISPs, as it mentions ISPs throttling other types of traffic. What does everyone think? --Eneufeld 00:31, 24 October 2006 (UTC)

Intro Paragraph
What does the acronym BW mean in the sentence: "Recent BW needs from Internet users have led ISPs to also shape their traffic by blocking certain ports." Is it suppose to mean bandwidth? 76.208.25.23 00:59, 19 December 2006 (UTC)

Overhaul
I've worked on projects relevant to traffic shaping so I can give a little of a behind-the-scenes look. I'll be trying to overhaul this page a bit in the next few days. Any questions and comments can reach me here on the talk page. Scriptedfate 22:07, 9 January 2007 (UTC)

Okay. That's part 2. Most of my experience is with the ISP angle, so I can't really comment on the Network/Information Theory parts of shaping. If you have any comments, bug me about 'em. Might need some wikification, as I'm sure I'm missing a bunch of relevant cross-links. Cheers! Scriptedfate 17:09, 10 January 2007 (UTC)


 * While I am not an expert in this area, I know a lot about closely related fields. So, I really can't judge the quality of your changes, but I have read them and I didn't see anthing that was obviously wrong or horribly biased.  There were a few more buzzwords than I like ("forward-thinking", "upsell", etc.), and I think some of the grammar could be clearer ("they care not/little if they", etc.), but it wasn't "bad" enough for me to consider changing.
 * I do think you might want to add a larger section on "critisisms", and in particular, relate it to net neutrality. Wrs1864 18:14, 10 January 2007 (UTC)


 * Along the lines of 'Traffic shaping, especially when applied to single protocols, is seen as a move away from Network neutrality which some view as a desirable goal'? The way I'm seeing it is that Network Neutrality involves specifically degrading the QoS of specific protocols, of some applications, or (in the more publicized case) of access to specific servers. In other words, if an ISP blocks or severely degrades the QoS of Skype (which is not as easy as it sounds), the speed of FTP (again, not so easy, and for similar reasons), or access to eBay (website ransom)... that would be anti-NetNeutrality. However, what if the ISP degrades the QoS of all VoIP, the speed of all file transfers, or access to all websites?
 * That's what I'd call being 'Categorically Neutral'. The ISP has chosen what services it doesn't want to encourage its subscribers to use. If subscribers continue to want to use those services, supply will meet the demand and an ISP will pop up offering the exact reverse. Or market analysts within the restrictive ISP will try to upsell (add extra revenue by offering an incremental service alongside an existing committed service. Sorry if it sounds buzzy) guaranteed QoS (or guaranteed non-throttling) on VoIP or what-have-you.
 * It's still not what I (as an avid BT user) would necessarily want... but I'm hoping for a time when I can pay $5 for HTTP and SMTP and an additional $9.95 for best-effort P2P... and who knows? Maybe even a $5.95 for a guaranteed 500Kbps five times a month for streaming movies over RTSP. I think a better model (better than claiming you're purchasing 3Mbps up to 60GB transfer each month when, in reality, you're sharing your 3Mbps with up to 25 other people (look up 'oversubscription' as a business model)) would make most of these problems go away.
 * But that's a bit idealistic, even for me.
 * There are criticisms of traffic shaping out there, especially those that go against net neutrality. I definitely don't want my Skype killed just to force users to switch to the in-house VoIP 'solution' (it has happened, but not in the western world (to my knowledge)). But if my BT is labelled 'best-effort' so that my VoIP 911 call can get through... *shrug*.
 * ... The reason I'm talking at this length is (aside from my inherent verbosity) so I can hear your thoughts on this. I don't want to write that section until I hear what you're thinking in terms of what I was thinking. Scriptedfate 21:51, 11 January 2007 (UTC)

I'm adding some detail from the networking angle and adding some technical references (af-tm-0056 particularly is quite a solid source for traffic management generally, part of the Anchorage Accord and widely implemented and deployed). TS is a specific technique, and I think ISPs are using it as a euphemism for "deliberately slowing down the customer's internet". Behind The Wall Of Sleep 08:06, 21 June 2007 (UTC)

References, References
I just noticed that, although this article was tagged as Unreferenced, there are references, so I conservatively changed the tag to Refimprove. Unfortunately, there are two different reference styles in the article, so I added the citation style tag.

Therefore, there are two things that need discussing: 1. Does this article have "enough" references? 2. What style should we standardize on? Ken g6 23:27, 4 June 2007 (UTC)


 * 1. No. I believe this article can use more references.  I'll try and mark the statements that need referencing.
 * 2. I vote for the footnote style for references. --Pchov 12:32, 10 June 2007 (UTC)


 * I agree on both counts. The subject matter is quite technical and well defined elsewhere.  I've added a couple of refs but I'm sure there are more.  Behind The Wall Of Sleep 08:08, 21 June 2007 (UTC)
 * Just read the ISPs/P2P sections and I agree that more citations are needed. We all know it's going on but to be encyclopaedic the contentious/controversial claims need specific sources. Also to make the referencing more consistent, should we convert all the external links to footnotes? Behind The Wall Of Sleep 09:43, 21 June 2007 (UTC)
 * I reverted the anonymous edit deleting Virgin Media from the ISP list. As far as I know, Virgin are still openly metering and policing user traffic (see ThinkBroadBand article here).  Whether this technically comprises shaping or not, I wouldn't like to say, but it's at least as close as most of the other ISPs on the list. Behind The Wall Of Sleep 10:25, 21 August 2007 (UTC)

Traffic Shaping vs QoS
How does Traffic Shaping compare/differ from QoS? —Preceding unsigned comment added by KevinTraver (talk • contribs) 01:04, 3 January 2008 (UTC)


 * No simple answer to that I don't think, but here's a shot. Traffic shaping is one of several technical measures comprising the discipline of Traffic Management. One of the most common goals of traffic management is achieving some desired set of QoS properties. So I guess TS is one of the things you might do to achieve QoS - the "might" is important, as it's certainly not the only measure, and it also not present in all cases.  Depending on your specific goals, traffic policing without TS might be adequate, or maybe just metering and billing would do the trick, but in general you need a whole stack of things to achieve QoS of which TS is just one. TM can also have other goals as well/instead, for example differentiating traffic purely to charge more for packets that are somehow "higher value".  Behind The Wall Of Sleep (talk) 10:40, 3 January 2008 (UTC)

Policing vs. Shaping
Leaky bucket is not an example for traffic shaping. It's traffic policing! Have a look here: http://www.cisco.com/warp/public/105/policevsshape.html#policingvsshaping In order to present a good overview about that area its important to work out the differences between shaping and policing. —Preceding unsigned comment added by 141.53.247.50 (talk) 14:18, August 28, 2007 (UTC)

I changed the title of this section from "Implementation" to "Policing vs. Shaping" because based on an initial read it seems this article doesn't handle the differences between the two consistently. For example the "Relationship to Traffic Management" section differentiates between the two and states that traffic policing is often labeled traffic shaping when referring to the practice of ISPs curtailing traffic that they consider unwanted or hurtful to their overall bandwidth/network requirements, yet further down in the "Three Types of Traffic" section, traffic shaping was cited as a tool used to stop undesired traffic altogether. I made a few word substitutions to try and reflect the role of traffic policing in this, rather than leave this inconsistent presentation. AbbyEdwards (talk) 00:04, 12 December 2007 (UTC)


 * I agree entirely with AbbyEdwards - and I think the area which needs to change is the ISPs section, as the standards and academic sources all distinguish the two carefully. The discussion about ISPs is really about traffic management generally, and strays into classification (all that stuff about BitTorrent for example) and traffic engineering and borders on Net neutrality. I know its a worthy topic, but I think it would be more appropriate in a separate article.  I've just spotted the article on bandwidth management, which I  think is both a better location for it and a better article.  I might try to eliminate the overlap. (btwos ducks in anticipation of angry reverts by disgruntled ISP customers) Behind The Wall Of Sleep (talk) 12:05, 28 December 2007 (UTC)


 * Well that obviously upset someone. The citations are gone and some selective linkspam is back.  And how on earth is traffic shaped without delay? It's either later or early, and early doesn't seem to be an option, right?  The idea that "proprietary" mechanisms that can shape traffic without delay might be a useful sales tool, but it's got no place in an encyclopaedia. Traffic shaping has been well-defined for at least a decade, and this article is now about something else. Where do we go from here? Behind The Wall Of Sleep (talk) 18:51, 30 December 2007 (UTC)
 * No edit comments or talk from the editor, so I've been "bold" and reverted it, on the basis that the new claims had no supporting references, and at the same time had deleted many that were already there. I will attempt to contribute positively as well by finding some more references for the technical definition.  I really didn't think the technical aspects of shaping were controversial at all - unlike the ISPs/net neutrality debate - so I thought expanding this section would make the article more grounded.  Behind The Wall Of Sleep (talk) 14:34, 3 January 2008 (UTC)

Major Internet Service Providers Applying Traffic Management
This section seems only to serve as a "name and shame" service, and has no supporting citations. The correlation that they are operating shaping rather than some other mode of traffic management (and hence its relevance to this article) seems at best to be guesswork. Most importantly, it seems to be accusing a list of highly visibly and influential companies of doing something their customers don't like. I'm sure you can tell by now that I don't think its encyclopaedic. If there is a good reason why it should be here (and I don't think a consumer action campaign, regardless of its merit, is such a good reason, see WP:SOAP) could everyone who adds a company please also add a specific citation showing that they are operating traffic shaping? I think we need something fairly solid, i.e. press release from the company concerned, or their terms and conditions of service; I don't think that disgruntled users on web forums and suchlike are going to cut it in this case. Behind The Wall Of Sleep (talk) 16:30, 15 January 2008 (UTC)

Application traffic discrimination versus per user traffic
Good distinction made between discriminating per-user traffic versus application protocol based discrimination towards the end of ISPs and Traffic Management section. Hope such efforts will lead to better network neutrality. Raanoo (talk) 12:26, 5 March 2008 (UTC)

TCP Window Shaping
Maybe we could move forward with this problem if we had an article (or section I suppose) on TCP Window Shaping written by someone who understands it (Bsdguru seems to have relevant information). It would be really useful to understand how it works, its pros and cons, interaction with non-TCP traffic, any queueing theoretic basis, citations etc. The current to-and-fro is making it hard to improve the article. Anyone? Behind The Wall Of Sleep (talk) 19:08, 5 January 2008 (UTC)

I'm a Network Administrator for an ISP, and I can't quite understand what TCP Window Shaping would have much to do with this article.

Basically the throughput is controlled by both the bandwidth and to a degree, the latency. As the latency gets higher, the window shaping needs to be greater.

As the load on an ISP gets higher, so does the latency generally.

Shaping a TCP window basically mean that as the load on an ISP is low (so low latency), then the customer will experience good throughput. As the load increases (higher latency), then the throughput decreases, therefore it's a mechanism to auto-throttle customers as the system gets busy.

It's a way to dynamically throttle customers based on the load on the system.

However it's very unreliable because it kills the function of the TCP Window in the first place, which is to allow maximum throughput over even very latent (300ms) links. If a customer tries to access a high latency server (say 150ms), and their Window has been shaped, they will have very poor (perhaps less than dialup) latency, depending on how big/small the ISP is shaping their window to.

It's a very poor way of throttling in all the implementations I've seen or tried to be sold. Sam, Mansfield, UK (talk) 21:28, 3 June 2008 (UTC)
 * Thanks Sam, this is consistent with my understanding of TCP Window Shaping. The individual who was advocating it an appropriate traffic shaping mechanism hasn't been back since, so I can only assume that there is no counterargument to your points, or any queueing theoretic foundation to support it.  I think we can keep it out of the article now. Behind The Wall Of Sleep (talk) 08:28, 10 June 2008 (UTC)

Benefits
The benefits section is in really bad shape. There are no citations and many statements are known to be false. Furthermore many of the supposed ways to achieve "benefits" are by committing acts that are of (at best) questionable legality. Somebody with quicker access to primary sources should revamp this section. The risk is that it reads a little too much like propaganda for ISPs that is detached from the technical reality. 173.10.180.101 (talk) 03:25, 7 January 2014 (UTC)

And probably elsewhere in this article, it needs a rewrite in light of bufferbloat.

The reason why people are seeing "congestive collapse" in wifi is more due to bufferbloat than anything else. Classification/fair queuing can help the situation some (move the pain to bulk data traffic), but an AQM is necessary to solve this problem.

In short, the root cause to many (particularly wifi) networks performance die under 100% load today is that excessive buffers get filled, inserting sometimes horrific amounts of delay, to the point of failure. Ironically, this is easiest to see in really radio quiet locations, so that random drops due to interference won't cause TCP to back off.

See http://gettys.wordpress.com and the bufferbloat articles in the January and February CACM. There are also some videos in YouTube worth looking at.

I'm in the middle of preparing blog entries about a new AQM algorithm, and how/why fair queuing and classification will also be necessary; they should appear next week.

JimGettys (talk) 17:33, 3 May 2012 (UTC)