Asymptotically optimal control of many-server heterogeneous service systems with H_{2}^{*} service times

Optimal control of many-server heterogenous service systems with service times that have a special hyper-exponential distribution, denoted by $H_{2}^{*}$ , which is a mixture of an exponential distribution and a unit point mass at 0, is considered. A?static priority policy that assigns priorities to server pools based on their service time distributions is proposed. This policy is shown to be asymptotically optimal in the many-server heavy-traffic regime in minimizing the total number of customers.     

Sojourn Times in the <Emphasis Type="Italic">M</Emphasis>/<Emphasis Type="Italic">PH</Emphasis>/1 Processor Sharing Queue

We give in this paper an algorithm to compute the sojourn time distribution in the processor sharing, single server queue with Poisson arrivals and phase type distributed service times. In a first step, we establish the differential system governing the conditional sojourn times probability distributions in this queue, given the number of customers in the different phases of the PH distribution at the arrival instant of a customer. This differential system is then solved by using a uniformization procedure and an exponential of matrix. The proposed algorithm precisely consists of computing this exponential with a controlled accuracy. This algorithm is then used in practical cases to investigate the impact of the variability of service times on sojourn times and the validity of the so-called reduced service rate (RSR) approximation, when service times in the different phases are highly dissymmetrical. For two-stage PH distributions, we give conjectures on the limiting behavior in terms of an M/M/1 PS queue and provide numerical illustrative examples.This revised version was published online in June 2005 with corrected coverdate     

Optimal choice of threshold in Two Level Processor Sharing

We analyze the Two Level Processor Sharing (TLPS) scheduling discipline with the hyper-exponential job size distribution and with the Poisson arrival process. TLPS is a convenient model to study the benefit of the file size based differentiation in TCP/IP networks. In the case of the hyper-exponential job size distribution with two phases, we find a closed form analytic expression for the expected sojourn time and an approximation for the optimal value of the threshold that minimizes the expected sojourn time. In the case of the hyper-exponential job size distribution with more than two phases, we derive a tight upper bound for the expected sojourn time conditioned on the job size. We show that when the variance of the job size distribution increases, the gain in system performance increases and the sensitivity to the choice of the threshold near its optimal value decreases. The work was supported by France Telecom R&D Grant “Modélisation et Gestion du Trafic Réseaux Internet” no. 46129414.     

Queues with Equally Heavy Sojourn Time and Service Requirement Distributions

For the G/G/1 queue with First-Come First-Served, it is well known that the tail of the sojourn time distribution is heavier than the tail of the service requirement distribution when the latter has a regularly varying tail. In contrast, for the M/G/1 queue with Processor Sharing, Zwart and Boxma [26] showed that under the same assumptions on the service requirement distribution, the two tails are equally heavy. By means of a probabilistic analysis we provide a new insightful proof of this result, allowing for the slightly weaker assumption of service requirement distributions with a tail of intermediate regular variation. The new approach allows us to also establish the tail equivalence for two other service disciplines: Foreground–Background Processor Sharing and Shortest Remaining Processing Time. The method can also be applied to more complicated models, for which no explicit formulas exist for (transforms of) the sojourn time distribution. One such model is the M/G/1 Processor Sharing queue with service that is subject to random interruptions. The latter model is of particular interest for the performance analysis of communication networks.     

Optimal control of a removable and non-reliable server in an infinite and a finite M/H2/1 queueing system

This paper deals with a single removable and non-reliable server in both an infinite and a finite queueing system with Poisson arrivals and two-type hyper-exponential distribution for the service times. The server may be turned on at arrival epochs or off at service completion epochs. Breakdown and repair times of the server are assumed to follow a negative exponential distribution. Conditions for a stable queueing system, that is steady-state, are provided. Cost models for both system capacities are respectively developed to determine the optimal operating policy numerically at minimum cost. This paper provides the minimum expected cost and the optimal operating policy based on assumed numerical values given to the system parameters, as well as to the cost elements. Sensitivity analysis is also investigated.     

Maximum entropy analysis to the N policy M/G/1 queueing system with a removable server

We study a single removable server in an M/G/1 queueing system operating under the N policy in steady-state. The server may be turned on at arrival epochs or off at departure epochs. Using the maximum entropy principle with several well-known constraints, we develop the approximate formulae for the probability distributions of the number of customers and the expected waiting time in the queue. We perform a comparative analysis between the approximate results with exact analytic results for three different service time distributions, exponential, 2-stage Erlang, and 2-stage hyper-exponential. The maximum entropy approximation approach is accurate enough for practical purposes. We demonstrate, through the maximum entropy principle results, that the N policy M/G/1 queueing system is sufficiently robust to the variations of service time distribution functions.     

Sojourn times in a processor sharing queue with service interruptions

We study the sojourn times of customers in an M/M/1 queue with the processor sharing service discipline and a server that is subject to breakdowns. The lengths of the breakdowns have a general distribution, whereas the on-periods are exponentially distributed. A branching process approach leads to a decomposition of the sojourn time, in which the components are independent of each other and can be investigated separately. We derive the Laplace–Stieltjes transform of the sojourn-time distribution in steady state, and show that the expected sojourn time is not proportional to the service requirement. In the heavy-traffic limit, the sojourn time conditioned on the service requirement and scaled by the traffic load is shown to be exponentially distributed. The results can be used for the performance analysis of elastic traffic in communication networks, in particular, the ABR service class in ATM networks, and best-effort services in IP networks.     

Sojourn time asymptotics in the M/G/1 processor sharing queue

We show for the M/G/1 processor sharing queue that the service time distribution is regularly varying of index -ν, ν non-integer, iff the sojourn time distribution is regularly varying of index -ν. This result is derived from a new expression for the Laplace–Stieltjes transform of the sojourn time distribution. That expression also leads to other new properties for the sojourn time distribution. We show how the moments of the sojourn time can be calculated recursively and prove that the kth moment of the sojourn time is finite iff the kth moment of the service time is finite. In addition, we give a short proof of a heavy traffic theorem for the sojourn time distribution, prove a heavy traffic theorem for the moments of the sojourn time, and study the properties of the heavy traffic limiting sojourn time distribution when the service time distribution is regularly varying. Explicit formulas and multiterm expansions are provided for the case that the service time has a Pareto distribution. This revised version was published online in June 2006 with corrections to the Cover Date.     

A decomposition theorem and related results for the discriminatory processor sharing queue

In this paper, we study a discriminatory processor sharing queue with Poisson arrivals,K classes and general service times. For this queue, we prove a decomposition theorem for the conditional sojourn time of a tagged customer given the service times and class affiliations of the customers present in the system when the tagged customer arrives. We show that this conditional sojourn time can be decomposed inton+1 components if there aren customers present when the tagged customer arrives. Further, we show that thesen+1 components can be obtained as a solution of a system of non-linear integral equations. These results generalize known results about theM/G/1 egalitarian processor sharing queue.     

M/G/1/MLPS compared to M/G/1/PS

Multilevel processor sharing scheduling disciplines have recently been resurrected in papers that focus on the differentiation between short and long TCP flows in the Internet. We prove that, for M/G/1 queues, such disciplines are better than the processor sharing discipline with respect to the mean delay whenever the hazard rate of the service time distribution is decreasing.     

Mean sojourn times for phase-type discriminatory processor sharing systems

In a discriminatory processor sharing (DPS) queueing model, each job (or customer) belongs to one out of finitely many classes. The arrival processes are Poisson. Classes differ with respect to arrival rates and service time distributions. Moreover, classes have different priority levels. All jobs present are served simultaneously but the fraction of the server’s capacity allocated to each one of them is proportional to their class priority parameter (while the total capacity is of course fixed).     

On first passage times of a hyper-exponential jump diffusion process

We investigate some important probabilistic properties relating to the first passage time of a hyper-exponential jump diffusion process, including its finiteness, expectation, conditional memorylessness, and conditional independence. Moreover, the joint distribution of the first passage time and the overshoot is studied from a primal-dual perspective.     

TheM/G/1 queue with processor sharing and its relation to a feedback queue

The central model of this paper is anM/M/1 queue with a general probabilistic feedback mechanism. When a customer completes his ith service, he departs from the system with probability 1–p(i) and he cycles back with probabilityp(i). The mean service time of each customer is the same for each cycle. We determine the joint distribution of the successive sojourn times of a tagged customer at his loops through the system. Subsequently we let the mean service time at each loop shrink to zero and the feedback probabilities approach one in such a way that the mean total required service time remains constant. The behaviour of the feedback queue then approaches that of anM/G/1 processor sharing queue, different choices of the feedback probabilities leading to different service time distributions in the processor sharing model. This is exploited to analyse the sojourn time distribution in theM/G/1 queue with processor sharing.Some variants are also considered, viz., anM/M/1 feedback queue with additional customers who are always present, and anM/G/1 processor sharing queue with feedback.     

A Study of Multiple Finite-Source Queueing Models

In this paper, the effects of the head-of-the-line priority and the first-come-first-served service disciplines and the variability of the service-time density functions on the mean values of performance measures of multiple finite-source queueing models are studied. In these models, with a single server, the times between the service completion and the next service requirement are exponentially distributed and the service times follow exponential, hypo-exponential or hyper-exponential probability density functions. The effects are compared with the known behaviour of multiple infinite-source queueing models.     

A heavy-traffic comparison of shared and segregated buffer schemes for queues with the head-of-line processor-sharing discipline

As network speeds increase and the data traffic becomes more diverse, the need arises for service disciplines that offer fair treatment to diverse applications, while efficiently using resources at high speeds. Disciplines that approximate round-robin or processor-sharing service per channel are well suited for data networks because, over a wide range of time scales, they allocate bandwidth fairly among channels without needing to distinguish between different types of applications. This study is among the few to address head-of-line processor sharing. In most previous models of processor-sharing disciplines, the system immediately serves any arriving message at a rate depending only on the number of messages in the system regardless of how these messages are distributed among the channels. This model is commonly called pure processor sharing. In our model, the server completes the work from a given channel at a rate depending on the number of other channels with work in the system. That is, the service rate depends on how messages are distributed among the channels, and only indirectly on the total number of messages in the system. In this paper, we contrast the buffer requirements of shared and non-shared buffer schemes, when both types of schemes provide head-of-the-line processor-sharing service among channels. We formulate the problem as a system of functions representing the cumulative input and cumulative lost (potential) output to parts of the queueing system and model the vector of input functions as a multi-dimensional Brownian motion. The resulting heavy-traffic approximations predict much larger benefits from sharing buffers than those predicted by pure processor sharing.     

Analysis of the M/M/1 Queue with Processor Sharing via Spectral Theory

We show in this paper that the computation of the distribution of the sojourn time of an arbitrary customer in a M/M/1 with the processor sharing discipline (abbreviated to M/M/1 PS queue) can be formulated as a spectral problem for a self-adjoint operator. This approach allows us to improve the existing results for this queue in two directions. First, the orthogonal structure underlying the M/M/1 PS queue is revealed. Second, an integral representation of the distribution of the sojourn time of a customer entering the system while there are n customers in service is obtained.     

On the distribution of exponential functionals for Lévy processes with jumps of rational transform

We derive explicit formulas for the Mellin transform and the distribution of the exponential functional for Lévy processes with rational Laplace exponent. This extends recent results by Cai and Kou [3] on the processes with hyper-exponential jumps.