Multiple Heterogeneous Servers Research Articles

Computation Offloading Strategy Optimization with Multiple Heterogeneous Servers in Mobile Edge Computing

Computation offloading from a user equipment (UE) to a mobile edge cloud (MEC) is an effective way to ease the computational burden of mobile devices, to improve the performance of mobile applications, to reduce the energy consumption and to extend the battery lifetime of mobile user equipments. In this paper, we consider computation offloading strategy optimization with multiple heterogeneous servers in mobile edge computing. Queueing models are established for a UE and multiple heterogeneous servers from different MECs, and the average task response time of the UE and each MEC server and the average response time of all offloadable and non-offloadable tasks generated on the UE are rigorously analyzed. Three multi-variable optimization problems are formulated, i.e., minimization of average response time with average power consumption constraint, minimization of average power consumption with average response time constraint, and minimization of cost-performance ratio, so that computation offloading strategy optimization, power-performance tradeoff, as well as power-time product can all be studied in the context of load balancing. An efficient numerical method (which consists of a series of fast numerical algorithms) is developed to solve the problems of minimization of average response time with average power consumption constraint, minimization of average power consumption with average response time constraint, and minimization of cost-performance ratio. Numerical examples and data are also demonstrated to show the effectiveness of our method and to show the power-performance tradeoff, the power-time product, and the impact of various parameters. To the best of the author's knowledge, this is the first work in the literature that analytically addresses computation offloading strategy optimization with multiple heterogeneous servers in mobile edge computing.

Everyone-Centric Heterogeneous Multi-Server Computation Offloading in ITS with Pervasive AI

The challenges of everyone-centric customized endto- end (E2E) service experience are spurring the worldwide interests on the sixth-generation (6G) networks. Nevertheless, in current intelligent transportation systems (ITS), fixed roadside units (RSUs) based multi-access edge computing (MEC) has the defects of poor flexibility, unbalanced utilization and high deployment cost, which cannot meet the diverse requirements for computation-intensive and latency-sensitive Internet of vehicles (IoV) services. In this article, we propose a novel MEC architecture with multiple heterogeneous servers and pervasive intelligence to provide computation offloading cooperatively for customized requirements of task vehicle users (TVUs). A potential game theory and federated deep reinforcement learning based approach, namely GT-FDRL, is proposed to decouple the offloading decision and resource allocation. First, a binary offloading model based on game theory is utilized to make offloading decisions. Secondly, horizontal federated learning is introduced to multiagent double deep Q-network (DDQN) to optimize the channel assignment and power allocation. Furthermore, the client-server architecture and federated aggregation are used to maintain the global model. A case study is presented and the simulation results demonstrate the effectiveness and robustness of our proposed approach.

Transient analysis of a Markovian queuing model with multiple-heterogeneous servers, and customers’ impatience

A Markovian queuing system with multiple-heterogeneous servers, reneging and retention of reneging customers is studied. The transient analysis of the model is performed using probability generating function technique. Important measures of performance like expected system size, average reneging rate and average retention rate are presented. The transient behavior of state probabilities is studied numerically. The transient behavior of performance measures is also studied. Further, a comparative analysis is performed which shows that the multiple-heterogeneous servers queuing model performs better than the multiple-homogeneous servers queuing model. Finally, the steady-state solution of the model is also obtained.

An evolutionary approach for optimal multi-objective resource allocation in distributed computing systems

Resource allocation in a distributed computing system is the process of allocating the workload across multiple computing resources to optimize the required performance criteria. In this article, a resource allocation problem that arises in a distributed system consisting of multiple heterogeneous servers is addressed. The problem is modeled as a multi-objective problem with two conflicting objectives: (a) to minimize the users’ expected response time and (b) to reduce the utilization imbalance between servers. To satisfy these objectives simultaneously, first, both the objectives are considered in an integrated manner, and an optimization problem is formulated. Second, the optimization problem is cast into a game-theoretic setting and modeled as a non-cooperative game, called a non-cooperative resource allocation game. Finally, to solve the game, a differential evolution-based co-evolutionary framework (DECEF) is proposed. To evaluate the performance of DECEF, a rigorous simulation study is carried out. Furthermore, to assess the relative performance of DECEF, it is compared against two existing approaches, from various aspects, including system utilization, system heterogeneity, and system size. The experimental results show that DECEF provides better system-wide performance while optimizing both the objectives.

Optimal Control of Service Systems with Heterogeneous Servers and Priority Customers

We study service systems with parallel servers and random customer arrivals, with a focus on the total waiting cost of customers. Using a Markov decision process (MDP) modeling approach, we analytically characterize the structures of the dynamic server assignment policies for two important systems, one consisting of multiple homogeneous servers and two classes of customers; and the other consisting of two heterogeneous servers and multiple classes of customers. Based on the obtained results, we propose a threshold-type heuristic policy for the generalized system consisting of multiple heterogeneous servers and multiple classes of customers. To design such a heuristic policy, we first develop techniques for the performance evaluation of general threshold-type policies with any given threshold values. We then construct a path to search for the optimal threshold values. Finally, we compare the performance of the best threshold-type heuristic policy with that of the optimal policy, and show that our proposed heuristic policy is computationally efficient yet generates great performance. We compare the system under our threshold-type dynamic server assignment policy with other commonly seen and simple systems (including the dedicated and the work-conserving flexible priority systems), and demonstrate the importance and usefulness of dynamic server assignment control, for service systems having multiple classes of customer arrivals. We also conduct sensitivity analysis to explore the impacts of various system configuration features (including the number of servers; service rate variation; number of customer classes; and system utilization) on the performance regarding waiting-cost minimization.

COOPER-SCHED: A Cooperative Scheduling Framework for Mobile Edge Computing with Expected Deadline Guarantee

While mobile edge computing (MEC) holds promise to enhance users' mobile experiences, building a scheduling framework to make full use of MEC capabilities is challenging. When involving quality of service (QoS) in MEC, the problem becomes even harder. In this work, we focus on QoS guaranteed scheduling in MEC with a cloudlet, which is a small cloud center deployed at the wireless access point (AP) to serve nearby mobile devices. There are multiple mobile devices (MDs) and each one is associated with a job to be offloaded to the AP and executed in the cloudlet. Each job is associated with a block of input data, an execution workload, and a QoS requirement, i.e., a time deadline that the job is expected to be completed before it. Our goal is to find an efficient schedule, which involves radio access network (RAN) allocation and job mapping on multiple heterogeneous servers, such that the number of jobs whose deadlines are satisfied is maximized. The problem is proved to be NP-hard. To solve the problem, we propose an extended marriage algorithm (EMA) by adapting the stable marriage game, for job mapping in the cloudlet. Based on this algorithm, we further implement a cooperative game based scheduling method COOPER-SCHED, which also involves RAN allocation. We perform extensive random experiments and compare it with three common heuristics in scheduling literature. The results show that COOPER-SCHED can find better schedules and shows more stability, i.e., suffering less impacts from RAN allocation, than others in MEC.

Optimal Task Dispatching on Multiple Heterogeneous Multiserver Systems with Dynamic Speed and Power Management

Cloud load balancing is the process of distributing workloads across multiple computing resources in a cloud environment. Load distribution in cloud computing systems is more challenging than in other systems. The purpose of the paper is to address the issue of optimal task dispatching on multiple heterogeneous multiserver systems with dynamic speed and power management. The main contributions of the paper are to solve three problems, i.e., the optimal task dispatching problems with minimized average task response time, minimized average power consumption, and minimized average cost-performance ratio, for multiple heterogeneous multiserver systems with dynamic $d$ -speed and power management. In our study, multiserver systems with dynamic speed and power management are modeled as queueing systems, so that fundamental performance and cost metrics such as the average task response time and the average power consumption can be obtained analytically. Our research problems are formulated as multi-variable optimization problems and solved numerically. To the best of our knowledge, this is the first work that addresses load distribution for performance optimization, power minimization, and cost-performance ratio optimization, collectively on multiple heterogeneous servers with dynamic speed and power management.

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems

We study the multi-resource allocation problem in cloud computing systems where the resource pool is constructed from a large number of heterogeneous servers, representing different points in the configuration space of resources such as processing, memory, and storage. We design a multi-resource allocation mechanism, called DRFH, that generalizes the notion of Dominant Resource Fairness (DRF) from a single server to multiple heterogeneous servers. DRFH provides a number of highly desirable properties. With DRFH, no user prefers the allocation of another user; no one can improve its allocation without decreasing that of the others; and more importantly, no coalition behavior of misreporting resource demands can benefit all its members. DRFH also ensures some level of service isolation among the users. As a direct application, we design a simple heuristic that implements DRFH in real-world systems. Large-scale simulations driven by Google cluster traces show that DRFH significantly outperforms the traditional slot-based scheduler, leading to much higher resource utilization with substantially shorter job completion times.

Managing performance and power consumption tradeoff for multiple heterogeneous servers in cloud computing

There are typically multiple heterogeneous servers providing various services in cloud computing. High power consumption of these servers increases the cost of running a data center. Thus, there is a problem of reducing the power cost with tolerable performance degradation. In this paper, we optimize the performance and power consumption tradeoff for multiple heterogeneous servers. We consider the following problems: (1) optimal job scheduling with fixed service rates; (2) joint optimal service speed scaling and job scheduling. For problem (1), we present the Karush-Kuhn-Tucker (KKT) conditions and provide a closed-form solution. For problem (2), both continuous speed scaling and discrete speed scaling are considered. In discrete speed scaling, the feasible service rates are discrete and bounded. We formulate the problem as an MINLP problem and propose a distributed algorithm by online value iteration, which has lower complexity than a centralized algorithm. Our approach provides an analytical way to manage the tradeoff between performance and power consumption. The simulation results show the gain of using speed scaling, and also prove the effectiveness and efficiency of the proposed algorithms.

OPTIMAL STATIC ASSIGNMENT AND ROUTING POLICIES FOR SERVICE CENTERS WITH CORRELATED TRAFFIC

A service center is a facility with multiple heterogeneous servers providing specialized service to multiple types of customers. An assignment policy specifies which server is enabled to serve which types of customer, and a routing policy specifies which server a customer will be routed to for service. Thus, a server can be enabled to serve many types of customer, and a customer may have many alternate servers who can serve him. This paper aims to provide decision models to determine optimal static assignment and routing policies, explicitly taking into account the stochastic fluctuations of demand along with the autocorrelations and cross-correlations of the different traffic streams. We consider several possible performance measures and formulate the optimization problem as a mixed integer nonlinear programming problem. We also develop an efficient heuristic algorithm to enhance scalability. Finally, we compare the different policies using the heuristic algorithms. We observe numerically that the routing policy tries to combine the negatively correlated traffic streams, and separate the positively correlated traffic streams.

Optimal power allocation among multiple heterogeneous servers in a data center

In a data center for cloud computing, there are typically multiple heterogeneous servers which provide services in different application domains. For such heterogeneous servers in a data center with different configurations for diversified applications and certain available power, there is a problem of allocating the power to the servers, such that the overall quality of service of the servers in the data center is optimized. We address power constrained performance optimization in a data center with multiple heterogeneous servers. We consider the problem of optimal power allocation among multiple heterogeneous servers, i.e., minimizing the average task response time of multiple heterogeneous computer systems with energy constraint. Each server is treated as a queueing system and the average task response time in a data center with multiple servers is formulated as a function of power allocations to the servers. The average task response time is minimized subjected to the constraint that the total effective power consumption of all the servers does not exceed a given power limit. We develop an algorithm to find the optimal solution and demonstrate numerical data. We also develop several closed-form heuristic solutions and show that they are very close to the optimal solution. Our approach provides an analytical way of studying the power-performance tradeoff at the data center level.

Analysis of job assignment with batch arrivals among heterogeneous servers

We revisit the problem of job assignment to multiple heterogeneous servers in parallel. The system under consideration, however, has a few unique features. Specifically, repair jobs arrive to the queueing system in batches according to a Poisson process. In addition, servers are heterogeneous and the service time distributions of the individual servers are general. The objective is to optimally assign each job within a batch arrival to minimize the long-run average number of jobs in the entire system. We focus on the class of static assignment policies where jobs are routed to servers upon arrival according to pre-determined probabilities. We solve the model analytically and derive the structural properties of the optimal static assignment. We show that when the traffic is below a certain threshold, it is better to not assign any jobs to slower servers. As traffic increases (either due to an increase in job arrival rate or batch size), more slower servers will be utilized. We give an explicit formula for computing the threshold. Finally we compare and evaluate the performance of the static assignment policy to two dynamic policies, specifically the shortest expected completion policy and the shortest queue policy.

Large deviations of Markovian polling models with applications to admission control

In this paper we consider large deviations and admission control problems for a discrete-time Markovian polling system. The system consists of two-parallel queues and multiple heterogeneous servers. The arrival process of each queue is a superposition of mutually independent Markovian on/off processes, and the multiple servers serve independently the two queues according to the so called Bernoulli service schedule. Using the large deviations techniques, we derive upper and lower bounds of the overflow probabilities, and then we present an admission control criterion by which different Quality of Service (QoS) requirements for the two queues are guaranteed.

The Alexandria Digital Library architecture

Since 1994, the Alexandria Digital Library Project has developed three prototype digital libraries for georeferenced information. This paper describes the most recent of these efforts, a three-tier client-server architecture that relies heavily on a middleware layer to present a single uniform set of interfaces to multiple heterogeneous servers. These standard interfaces, all of which are implemented in HTTP, support session management, collection discovery and evaluation, metadata searching, metadata retrieval, and online holding retrieval. An XML-based metadata encoding scheme and a simple Boolean query language have also been developed. The architecture described by these interfaces has been implemented at UCSB.

File allocation with balanced response time in a distributed multi-server information system

As information systems move from centralized to distributed systems, the distribution of the data files among the servers becomes a major issue. This paper reports on the results of a heuristic algorithm that is used to distribute the files among multiple heterogeneous servers interconnected by a fast network. The objective of our algorithm is to minimize the differences between response times of the servers (or in other words, to minimize the response imbalance factor). The servers are modeled using M/M/1 queues. In this paper, we first describe the problem, then the algorithm is presented and its bounds on the imbalance factor analyzed. The characteristics of our algorithm are also studied through simulation. Results are compared with the algorithms proposed by other researchers, which show a considerable improvement in the imbalance factor and only a slight increase in the average response time of the system.

A stochastic recirculation system with random access

A model for a stochastic recirculation system with randomly accessed multiple heterogeneous servers, no waiting rooms, and exponentially-distributed service times is provided. In this system the units are assigned to one of the servers upon arrival by random mechanism. Units which find all servers busy recirculate and combine with the incoming arrivals and join those already in the system to initiate the next cycle. The equilibrium behavior of the internal and external stochastic processes of the system is analyzed using a two parameter approximation. A simulation model is also developed and its behavior is compared against the analytical model at the steady state. The model with randomly-accessed servers is compared to a single server model already established in the literature. The performance of the model is then examined for a wide range of parameter values to obtain conclusions about its optimal performances.

Queuing with reneging and multiple heterogeneous servers

AbstractA Poisson stream of items arrive at a multiple parallel serving facility consisting of s heterogeneous servers. The service time density functions are all negative exponential and may all have different means. An arrival balks (refuses to enter) if the queue size is equal to N. If the queue size is less than N, the arrival enters the queue and is served on a first‐come, first‐served basis unless it reneges (leaves the queue). Each entering item has a definite length of time it will wait in queue before reneging. This time is a random variable with negative exponential density function. If on arrival more than one server is free, the server to be used is chosen at random. The results obtained are for steady‐state and include (1) state probabilities; (2) the probability of j or more in system, j=s, s+I,…, s+N; (3) mean number in queue and system and mean number of busy servers; (4) probability that an arrival acquires service; (5) mean rate of loss due to balking and reneging; and (6) density functions and mean values of waiting times in queue for those who are served, those who renege, and all who enter. Certain important special cases are derived from the preceding, namely, (1) unbounded queues, (2) homogeneous servers, (3) no reneging, (4) no reneging and unbounded queues, and (5) single server. Finally, a comparison is made of a heterogeneous system with an equivalent homogeneous system.