Talk:Quality of service

UNIX System Administration Handbook

 * I think this article may benefit from information contained in this book. What happened is in the late 90s, Microsoft released an operating system whose TCP/IP stack labelled all packets "important", completely disregarding the above agreement in the course of doing so. This meant that interactive data from Unix/Linux boxes were being dropped if any Windows traffic was on the pipe. The other operating systems responded by turning off the DSCP feature making the use of these bits moot. As long as all these Window installation are sitting somewhere in the net, I don't think this concept will ever take off.


 * May be MPLS will replace it, as it looks like some people are happy to see a tied Internet. Personally, I don't see the technical need for it considering the amount of dark fibre around, but bussiness wise (Read profit) there may be a case. Anyway time will tell


 * If someone has some time to spare, pick the above text and verify my assertion. It has a good political story that is relevant to this article

DiffServ bits "honoured" in peering connections?
The text: "These bits were later re-defined as DiffServ Code Points (DSCP) and are largely honored in peered links on the modern Internet." seems to indicate that routers at Internet exchanges etc. use the DiffServ bits when prioritising packets to be sent. This did not strike me as correct, and later it says clearly that DiffServ is not widely deployed on the Internet. Can someone with direct knowledge of peering connections between ISPs and of core routers at Internet exchanges clarify or correct this? Robin Whittle 07:20, 9 February 2007 (UTC)


 * Typically not and the article no longer makes this claim. ~Kvng (talk) 21:28, 6 June 2022 (UTC)

QoS Problems section
This one-sentence “section” seems to be highly biased and grossly oversimplified and misleading by insinuating that QoS is moot. The entire section generalizes that QoS isn’t economically sound and purports to be from an "Internet 2 working group" when it's actually a 2002 paper written by Stanislav Shalunov and Benjamin Teitelbaum. Of the two references in this section, one is merely an article touting the work of Shalunov and Teitelbaum and the other reference points to the paper itself.

The title of these two gentleman's work "Why Premium IP Service Has Not Deployed (and Probably Never Will)" and the conclusions are simply opinion and they're not supported by the body of the paper. The paper makes sweeping generalizations about QoS being unnecessary on the Internet in general yet the paper admits that:


 * "Internet2 networks are generally well-provisioned and almost always lightly loaded. Packet loss and jitter experienced by best-effort traffic on Internet2 paths is almost always zero or is due to non-congestive causes"


 * “This is especially true when there is no per-bit charge for Internet traffic, as is the case within Internet2. Without pricing disincentives, individual users can very significantly and very suddenly affect network utilization.”

Shalunov and Teitelbaum are clearly admitting that the Internet 2 which is a well provisioned and metered network is uniquely less of a candidate for QoS. This is not applicable to the lesser provisioned and unmetered Internet in general. But even as congestion- and jitter-free as the Internet 2 generally is, Shalunov and Teitelbaum still admit that there are situations where QoS is necessary on the Internet 2 when they wrote:


 * “Premium service is about *guaranteeing* service quality. In essence, it is about removing a component of unreliability from the system -- the probability that a network transaction fails because of network congestion.  Although typical performance may be perfect, there would be considerable value in being able to assure that important sessions receive perfect network performance.  Who wants the possibility that their important conference calls are disconnected or suddenly deteriorate in quality?  Who wants a surgeon operating through robotic means on a different continent to experience IP packet loss artifacts?”

Shalunov and Teitelbaum go beyond the scope of the Internet 2 and offer some rather biased opinions by concluding:


 * “In the U.S. today, the price of network capacity is low and falling (with the notable exception of residential and rural access) and the apparent one-time and recurring costs of Premium are high and rising (with interface speeds). In most bandwidth markets important to network-based research, it is cheaper to buy more capacity and to provide everybody with excellent service than it is to mess with QoS.  In those few places where network upgrades are not practical, QoS deployment is usually even harder (due to the high cost of QoS-capable routers and clueful network engineers).”

Residential and rural access is quite a large exception to the paper's conclusion that increasing bandwidth is cheap. In fact it’s probably the entire broadband industry. The claim that QoS-capable routers and clueful network engineers is too expensive is also proven false by the existence of widely deployed Fiber to the Node (FTTN) networks which run QoS to ensure that their IPTV service has guaranteed bandwidth and low jitter. Lastly, Shalunov and Teitelbaum concede that ad hoc QoS solutions work well.


 * “In the very few cases where the demand, the money, and the clue are present, but the bandwidth is lacking, ad hoc approaches that prioritize one or two important applications over a single congested link often work well. In this climate, the case for a global, inter domain Premium service is dubious.”

In conclusion, this one-sentence section should either be deleted or revised. A more appropriate conclusion that should be drawn from the Shalunov and Teitelbaum paper is that a large-scale Resource Reservation Protocol implementation across the entire Internet or Internet 2 is probably infeasible but point solutions that implement Differentiated Services is probably the way to go. GeorgeOu 05:24, 21 September 2008

There seems to be some profound confusion in this section.

1. QoS mechanisms only need to go into action when there is scarce capacity and various communications flows are competing. So if there is enough capacity, QoS is not necessary.
 * This is not true if the goal of your QoS mechanism is to minimize jitter.


 * You don't get much jitter if you have a largely empty network though; and if your network isn't largely empty, then your network is already full due to exponential traffic growth.WolfKeeper 14:21, 4 August 2007 (UTC)
 * I think you're wrong here. Packet collisions at routers are very common, even when networks are largely empty. This is an example of the birthday paradox, except that we don't have birthdays colliding. We instead have packets with similar arrival times colliding. Mc6809e 09:01, 7 August 2007 (UTC)
 * That kind of jitter is only a small cause of delay. The major causes of jitter are when the queues tend to fill, which happens on a busy network.WolfKeeper 09:15, 7 August 2007 (UTC)
 * Filling queues is only evidence that there are collisions. But full queues are not the cause of jitter. It's entirely possible to have nearly full queues and no jitter at all. If packets all arrive at nearly the same time, but always in the same order, then there is no jitter. Just an increase in latency for those packets arriving latest. So full queues increase latency (without QoS), but don't necessarily cause jitter. Mc6809e 19:07, 7 August 2007 (PST)
 * Agreed Mc6809e. I would add that if the different streams are offset either by random luck or by scheduling, then there's zero latency and zero jitter added. GeorgeOu 19:03, 21 September 2008
 * Wolfkeeper, it's not true that busy networks cause filled up queues and lightly networks don't have filled up queues. You could have traffic streams that operate at 90% capacity but have no more than 1 or 2 packets in the transmit queue if the stream is isochronous and the packets are evenly spaced out.  This is precisely why IPTV streams that take most of the capacity on an FTTN link cause zero measurable increase in ping times.  On the other hand, a 20% load from multiple peer-to-peer TCP flows can cause lots of jitter and very deep queue depths.  So the idea that abundant capacity and low utilization levels can substitute for QoS is fundamentally misguided.GeorgeOu 18:56, 21 September 2008

2. QoS mechanisms only seem to work when the network is almost full and not fully full. The result is QoS only adds a couple of percentage points of extra capacity under the right circumstances.
 * Here it is important to carefully define what is meant by capacity; if there is a bottleneck in the network running at full capacity, QoS mechanisms cannot increase this capacity.
 * Good point. You have to ask: capacity for what? VoIP calls? Bulk data transfer? A saturated 100 Mbps link without QoS can have zero VoIP capacity if internal buffers are large and holding packets for more than 300 milliseconds. Adding QoS can allow the link to go from being able to carry zero VoIP calls to being able to carry literally thousands. Mc6809e 09:01, 7 August 2007 (UTC)

5. It is cheaper to upgrade to higher capacity lines or equipment.
 * This would depend on the scenario and what the goal for the QoS mechanisms is.

6. Traffic growth of IP-traffic is up to 100% per year in the western world. This means that the benefits of QoS are often short-lived. It gets an operator through the month, not through to next year.
 * This makes no sense. It seems to imply that the operators can somehow gain capacity through QoS mechanisms.
 * Jitter due to network contention can prevent VOIP from working, but deploying QoS minimises these problems, but only for a few months.WolfKeeper 14:21, 4 August 2007 (UTC)
 * QoS will fix things like VoIP for a very long time. Voice calls are typically very low bits/second transmissions. Increasing bandwidth helps only because it clears other traffic quickly enough to keep buffers relatively empty and allows VoIP packets to quickly move through the router. But QoS mechanisms can allow this low bandwidth traffic to bypass bulk traffic in queues. The question for the ISP is whether or not QoS is more expensive than increasing bandwidth. Mc6809e 20:20, 7 August 2007 (UTC)

7. QoS needs management and constant attention to determine that shortages don't occur. Overengineering the network results in less attention required.
 * This would depend entirely on the QoS mechanism. It is implying that it's theoretically impossible to develop a QoS mechanism that requires little "attention", such claims must be sourced.
 * Agreed. Most routers will let you use source port as a means of prioritization. It's easy to setup. Besides, we don't know the complete economics of QoS. It may be that QoS management is less expensive than increasing bytes/second in some cases. Mc6809e 20:34, 7 August 2007 (UTC)

9. Getting QoS to work over the edges of multiple networks requires management and agreements of how to prioritize traffic between those networks. It's often easier to add extra capacity to interconnects.
 * The first sentence is true, the second part might be true under some scenarios but is not true in the general case.

10. Where it is possible to make guarantees on capacity it's generally easier not to determine priority per flow, but to give the customer a guaranteed amount of capacity between two points. This allows the customer to make decisions on priority. This is the basis of most MPLS implementations. --Teemuk 09:30, 14 May 2007 (UTC)
 * This again boils down to QoS mechanisms vs. more capacity. As before it depends entirely on the scenario and goals.
 * I agree. There is so much wrong, and uncited material I might add, that it's tempting to remove the whole section.


 * The material is uncited and will be deleted. Too much of it is just flat-out wrong. Mc6809e 20:34, 7 August 2007 (UTC)


 * There is no longer a Problems section. There is a Limitations section and I've reviewed it and didn't find any egregious issues. ~Kvng (talk) 21:34, 6 June 2022 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified 10 one external links on Quality of service. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20070111015452/http://oldwww.com.dtu.dk/teletraffic/handbook/telenook.pdf to http://www.com.dtu.dk/teletraffic/handbook/telenook.pdf
 * Added archive http://web.archive.org/web/20081219124325/http://archive.computerhistory.org/lectures/an_eveninig_with_robert_kahn.lecture.2007.01.09.wmv to http://archive.computerhistory.org/lectures/an_eveninig_with_robert_kahn.lecture.2007.01.09.wmv
 * Added archive http://web.archive.org/web/20110113060748/http://www.tmforum.org/ipsphere to http://www.tmforum.org/ipsphere/
 * Added archive http://web.archive.org/web/20110124185840/http://www.apps.ietf.org:80/rfc/rfc2474.html to http://www.apps.ietf.org/rfc/rfc2474.html
 * Added archive http://web.archive.org/web/20110126005909/http://www.apps.ietf.org:80/rfc/rfc2205.html to http://www.apps.ietf.org/rfc/rfc2205.html
 * Added archive http://web.archive.org/web/20120321203319/http://www.apps.ietf.org/rfc/rfc2990.html to http://www.apps.ietf.org/rfc/rfc2990.html
 * Added archive http://web.archive.org/web/20120115073105/http://www.apps.ietf.org:80/rfc/rfc3714.html to http://www.apps.ietf.org/rfc/rfc3714.html
 * Added archive http://web.archive.org/web/20110825144115/http://www.apps.ietf.org/rfc/rfc1633.html to http://www.apps.ietf.org/rfc/rfc1633.html
 * Added archive http://web.archive.org/web/20110125183757/http://www.apps.ietf.org:80/rfc/rfc2475.html to http://www.apps.ietf.org/rfc/rfc2475.html
 * Added archive http://web.archive.org/web/20120321203606/http://www.apps.ietf.org/rfc/rfc3209.html to http://www.apps.ietf.org/rfc/rfc3209.html

When you have finished reviewing my changes, please set the checked parameter below to true or failed to let others know (documentation at ).

Cheers.— InternetArchiveBot  (Report bug) 12:30, 21 July 2016 (UTC)