Proportional delay differentiation provision by bandwidth adaptation of class‐based queue scheduling

Abstract

SUMMARYSupporting quality of service (QoS) over the Internet is a very important issue and many mechanisms have already been devised or are under way towards achieving this goal. One of the most important approaches is the so-called Differentiated Services (DiffServ) architecture, which provides a scalable mechanism for QoS support in a TCP/IP network. The main concept underlying DiffServ is the aggregation of traffic flows at an ingress (or egress) point of a network and the marking of the IP packets of each traffic flow according to several classification criteria. Diffserv is classified under two taxonomies: the absolute and the relative. In absolute DiffServ architecture, an admission control scheme is utilized to provide QoS as absolute bounds of specific QoS parameters. The relative DiffServ model offers also QoS guarantees per class but in reference to the guarantees given to the other classes defined. In this paper, relative proportional delay differentiation is achieved based on class-based queueing (CBQ) scheduler. Specifically, the service rates allocated to the classes of a CBQ scheduler are frequently adjusted in order to obtain relative delay spacing among them. The model presented can also be exploited in order to meet absolute delay constraints in conjunction with relative delay differentiation provision. Simulation experiments verify that our model can attain relative as well as absolute delay differentiation provided that the preconditions posed are satisfied.

Fair bandwidth sharing and delay differentiation: Joint packet scheduling with buffer management

“…Delay and bandwidth are two important QoS metrics considered in the model. The algorithms for proportional delay differentiation (PDD) consider lossless and work-conserving packet scheduling [4][5][6]11,13,14,17,18]. When the overall workload of classes is close to or exceeds the link bandwidth capacity, the algorithms for proportional bandwidth differentiation aim to enforce that the ratio of the loss rates of two classes be proportional to the ratios of their differentiation parameters [3,7,21,23].…”

“…Due to its per-class stateless processing, the DiffServ architecture exhibits good scalability. Its provisioning is an active research topic [5,7,9,10,17,[21][22][23][24][25].…”

Combining Response Surface Methodology with Numerical Methods for Optimization of Markovian Models

“…In particular, we study the robustness of the algorithm for a wide range of initial parameter settings. As a second and more complex application example, we consider the optimization of a class-based queueing system [10], [17]. Class-based queueing (CBQ) is a "per hop" packet-scheduling mechanism that provides differentiated service to traffic flows of different types and is used as part of the differentiated service architecture (DiffServ) [13].…”

Combining Response Surface Methodology with Numerical Models for Optimization of Class-Based Queueing Systems

“…Since φ is only implicitly represented as a stochastic model, or in other words as a black box, only those optimization methods that do not exploit the structure of φ can be applied. To solve that optimization (6) Approximate response surface function in the local region with center point c old and at least k design points by a first-order linear regression metamodel (7) Test the first-order model for goodness-of-fit according to the adjusted coefficient of determination (8) IF first-order model is adequate THEN DO (9) determine direction of steepest ascent/descent and step size (10) REPEAT (11) go one step in direction of steepest ascent/descent and determine the response for the new input parameters via a single evaluation of the stochastic model (12) UNTIL new response results in no further improvement (13) c new := input parameters of last improvement of the response (14) IF c new = c old THEN set new half-width ω := ω/2 (15) ELSE DO (16) IF ω < 4·ω stop THEN DO (17) Approximate the response surface in the local region with center point c old by a second-order linear regression metamodel (18) Test [15]. RSM is an optimization method that is able to identify a (local) optimum.…”

“…The approach trades computation time for accuracy. As an application example, we consider the optimization of a class-based queueing system [7], [13]. Class-based queueing (CBQ) is a "per hop" packet-scheduling mechanism that provides differentiated service to traffic flows of different types and is used as part of the differentiated service architecture (DiffServ) [9].…”