Friday, November 21, 2008

How Important Is QoS (Quality Of Service) To Deciding A Bandwidth Solution?

When going through the decision process for a bandwidth solution .... for example T1 or DS3 bandwidth .... QoS is AND should be a major consideration.

What Does (and should) QOS Mean To You?

Quality of Service (QoS) Definition:

In computer/telecom networks, a traffic engineering term. Refers to resource reservation control mechanisms, not service quality per se. Ability to prioritize applications, users, data flows, or to guarantee data flow performance.

But you knew/know that.

Alternative to QoS is high quality "best-effort" network communication, by over-providing capacity, such that it is sufficient for expected peak traffic loads. But high QoS often gets confused with high level performance, or achieved service quality (e.g., high bit rate, low latency, low bit error probability, and other arcane measures).

From a user perspective, however, QoS in telephony and streaming video, it is necessary to try to measure the quality that the user actually experiences, called "Quality of Experience" (or QoE), based on user perception or their level of satisfaction. Thus QoS needs to track subscriber satisfaction, which obviously is subjective and varies on a number of factors at the user end.

And when combined with experience, you get a sort of hybrid measure that can be called "Quality of Service Experience" (or QoSE), which tries to measure user experience relative to delivered quality, regardless of bit rates or other arcane measures. None of those matter if the user is not satisfied.

That's what QoS means to me.

I believe QoS is meant to measure the performance of the application delivered to the end user. If we references QoE when discussing QoS, we can now apply an experience based model (QoE) to the traditional network measures of service delivery quality (QoS). The addition of QoE to traditional QoS measures allows us to restate my 1st line to “QoE is meant to measure the performance of the application experienced by the end user”.

Since QoS measures the performance of the application, the measure of QoS varies depending on the application (Voice, VoIP, Video, Data, etc…). I believe that while QoS provides a good indication of the performance of the delivery of the application, the measure should be used hand-in-hand with some type of experience based model when available, such as QoE.

I guess QoS in in the eye of the beholder. It depends on the application, the network, the level of traffic and (not sure about this) the phase of the moon. Yet another measure of QoS is the MOS or Mean Opinion Score used in an attempt to quantize subjectivity in the Voice application space. Then there is the ITU-Ts Emodel...(so how was that telephone call for you?)

In academia thousand of research papers have been written and argued about. In the industry we have seen large numbers of IEEE specifications regarding QoS which all rely upon an assumption: Layer a software / CPU centric fix-it mechanism over best-effort IP comprising one or another packet prioritization scheme and you will have QoS. The industry calls this soft QoS.

None of these schemes are suited to traffic crossing multiple operator domains. MPLS over IP is a core technology, what we need for end-to-end QoS is an end-to-end solution. Hard QoS.

So how about starting off with a clean slate design for global QoS. My starter for ten (for the sake of avoiding unnecessary complexity) would be lets insist on only two classes of QoS: best-effort and best-quality. Best-effort needs no introduction - thats what we are stuck with in today's Internet - no guarantees whatsoever, you get what you get, take or leave it. Best-quality should by definition enable IP networks for all the latency critical applications such as voice, videoconferencing, gaming database replication...on a global scale.

And thats where the next few killer apps might come from if we can implement two classes of QoS in a globally scalable, self managing manner.

Labels: , , , , , ,


Post a Comment

Links to this post:

Create a Link

<< Home