Communication-Aware Scheduling of Serial Tasks for Dispersed Computing

Mobile opportunistic networks are characterized by intermittent connectivity and practically no end-to-end connection. In these kinds of network, network capacity and throughput largely depends packet replication. As a result buffer space tends to quickly deplete which may result in severe data loss due to packet drop. To the best of our knowledge, the two important phenomenon of traffic management viz., jitter and packet loss have evaded the minds of researchers in this field and hence we study the problem of buffer management and packet scheduling in such networks. We propose a fair queue scheduling policy where packets are sorted based on total waiting time, i.e. sum of waiting times over all the previous hops, before being transferred over a network link. We analytically prove that scheduling packet transmission as above reduces overall waiting time which leads to lower delay. We also propose a source initiated congestion aware buffer management scheme which helps in reducing number of packet loss significantly. The proposed scheduling and buffer management policy is simulated using real world (Cambridge iMote traces) datasets and compared with other existing scheduling policies. Results show that that proposed scheme has lower delay and lower packet drop compared to other scheduling schemes.

We study the scheduling of computational tasks on one local processor and one remote processor with communication delay. This problem has important application in cloud computing. Although the communication time to transmit a task can be inferred from the known data size of the task and the transmission bandwidth, the processing time of the task is generally unknown until it is processed to completion. Given a set of independent tasks with unknown processing times, we propose a Semi-online Partitioning and Communication (SPaC) algorithm, to jointly select the subset of tasks to be offloaded to the remote processor and their order of transmission, with an aim to minimize the overall makespan to process all tasks. Even though the offline version of this problem, with a priori known processing times, is NP-hard, we show that the proposed semi-online algorithm achieves a small competitive ratio when the communication times of tasks are smaller than their processing times at the remote processor. For general communication times, we use simulation to demonstrate that SPaC outperforms online list scheduling and performs comparably well with known offline heuristics.

Cloud manufacturing (CMfg) converts the traditional manufacturing system into an Internet-of-things-enabled (IoT-enabled) manufacturing system, where both manufacturing and computational tasks must be scheduled among distributed and heterogeneous resources. Deep reinforcement learning (DRL) has recently become a promising idea for task scheduling in CMfg. However, existing DRL-based methods depend heavily on problem-specific reward engineering and struggle to represent hybrid decision variables. To this end, this paper proposed the sparse-reward deep reinforcement learning (SDRL) method to solve the hybrid task scheduling problem in CMfg. First, the hybrid task scheduling model in CMfg is constructed to minimize the makespan. We reformulate the studied problem as a partially observable Markov decision process (POMDP). Then, the objective hindsight experience replay (objective HER) mechanism is proposed to alleviate the sparse reward issue, through which the scheduling policy can be effectively trained without problem-specific reward engineering. The continuous action space is defined to represent hybrid decision variables, and the implicit action-selection mapping is utilized to alleviate the boundary effect. Numerical experiments validated the effectiveness and superiority of our method compared to eleven popular scheduling algorithms including evolutionary algorithms and DRL. Compared to mainstream DRL scheduling methods, the proposed SDRL outperforms the second-best one at most by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$23.6\%$</tex-math> </inline-formula> regarding generalization, and a scheduling solution can be generated in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$0.5$</tex-math> </inline-formula> seconds. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —With the intelligentization of the CMfg platform, hybrid tasks, including manufacturing and computational tasks, need to be scheduled simultaneously. However, this hybrid task scheduling problem is rarely considered by existing works. DRL exhibits many benefits in addressing scheduling problems, but the strong dependency on problem-specific reward engineering limits its application. Additionally, most DRL-based scheduling algorithms are discrete-action DRL, restricting their capacity to effectively represent hybrid decision variables. The studied problem originates from the CMfg platform, but the proposed method holds potential for broader application. The scheduling framework and the POMDP modeling can be applied to similar problems, including hybrid, manufacturing, or computational task scheduling problems. The proposed objective HER serves as a general approach to addressing challenges associated with sparse rewards, which can be extended to diverse combinatorial optimization problems aimed at optimizing an objective. We will open-source our codes to help others to apply the method to other fields.

In multiple-user communications, the bursty nature of the packet arrival times cannot be divorced from the analysis of the transmission process. However, in traditional information theory the random arrival times are smoothed out by appropriated source coding and no consideration is made for the end-to-end delay. In this thesis, using tools from network theory, we investigate simple models that consider the end-to-end delay and/or the variability of the packet arrivals as important parameters, while staying in a information theoretic framework. First, we simplify the problem and focus on the transmission of a bursty source over a single-user channel. We introduce a new measure of channel features that enable us to incorporate the possibility to code among several packets in a scheduling problem. In this setup, we look for policies that minimize the average packet delay. Assuming that the packets are independent and sufficiently large to perform capacity achieving coding, we then consider the problem of allocating rates among a finite number of users communicating through a multiple-user channel. Following the previous work in the context of multiple-access channel, we formulate a scheduling problem in which the rate of each user is chosen from the capacity region of the multiple-user channel. Here, the goal is to find a scheduling policy that minimizes the sum of the transmitter queue lengths, such a policy is called delay optimal. In particular settings, for the additive Gaussian multiple-access channel we show the delay optimality of the Longer Queue Higher Rate policy introduced by Yeh. And, when the users communicate through a symmetric broadcast channel, we propose and show the delay optimality of a Best User Highest Possible Rate policy, among a large class of admissible policies. Finally, in the last part of this thesis, we look at the multiple-user channel coding problem from the perspective of the receivers. By measuring the transmission rates at the receivers, we are able to define variable length codes and to characterize the region of achievable rates when the receivers can decode their intended messages at different instants of time. For the two-user degraded broadcast channel and for the two-user multiple-access channel, we show that the gain in using variable length codes essentially comes from the possibility for the receivers to decode the transmitted messages in non-overlapping periods of time.

We consider in this paper two types of queueing systems which operate under a specific and fixed scheduling policy. The first system consists of a single server and several buffers in which arriving jobs are stored. We assume that arriving parts may require several stages of processing in which case each stage corresponds to a different buffer. The second system is a communication type queueing network given by a graph. The arriving jobs (packets) request a simple path along which they need to be processed. In both models the jobs arrive in a completely deterministic fashion: the interarrival times are fixed and known. All the processing times are also deterministic. A scheduling policy specifies a rule using which arriving parts are processed in the queueing system. A scheduling policy is defined to be stable if there is a finite uniform upper bound on the total number of parts in the system at all times. A necessary condition for stability of any work conserving policy is that the traffic intensity of the station (of each link in the graph in the communication model) is not bigger than one. Many results have demonstrated that this condition is not sufficient for stability. The results were obtained primarily in the context of stochastic networks ([24],[20],[5],[6],[10]), deterministic fluid networks ([6],[9],[7],[8],[2]), deterministic adversarial networks ([4],[1],[14],[12]). One of the earliest result in the area were obtained by Rybko and Stolyar [24] and Lu and Kumar [20]. They showed that a simple priority policy can lead to instability in some queueing networks. Bramson [5] and Seidman [25] showed that even FIFO policy can be unstable in stochastic networks. Instability of FIFO was later demonstrated in an adversarial queueing setting by Andrews et. al. [1]. Dai [6] established that stability of a fluid deterministic queueing network implies stability of a stochastic queueing network. A similar result was established by Gamarnik [12], which connects stability of fluid and adversarial queueing networks. A complete characterization of two-station fluid networks which are stable under any work conserving policy was established by Bertsimas, Gamarnik and Tsitsiklis [2] and Dai and Vande Vate [7]. Goel [14] constructed a complete characterization of adversarial queueing networks which are stable under the usual load condition. The result is extended by Gamarnik [13]. Motivated by a queueing network model stability of homogeneous random walks in a nonnegative orthant was considered in several papers: Malyshev [21], Menshikov [23], Fayolle [11], Ignatyuk and Malyshev [18], Malyshev [22]. Such random walks Q ( t ), t = 0, 1, 2,¡­ have Z + d as a state space ( Z + is the set of nonnegative integers). The transition vectors ¦¤ have deterministically bounded length in max norm and the transition probability p (¦«, ¦¤) along the direction ¦¤ depends only on the face ¦« that the random walk is currently on (the transition probabilities depend only on which coordinates of the current state are positive and which are zero). Such a random walk is defined to be stable if it is positive recurrent. We will also consider deterministic walks, for which p (¦«, ¦¤) is always zero or one (the transition deterministically depends on the face that the walk is currently on). A complete characterization of stable homogeneous random walks in Z + d for d ¡Ü 4 was obtained in Malyshev [21], Menshikov [23] and Ignatyuk and Malyshev [18], but no extension of this classification to higher dimensions has been obtained. Malyshev in [22] establishes a connection between homogeneous random walks and general dynamical systems on compact manifolds and shows that the difficulty of classifying stable random walks is of the same nature as the difficulty of understanding the dynamics of these dynamical systems. Specifically, the complicated dynamics precludes obtaining classification of stable random walks for d = 5. Thus despite many efforts the classification of stable random walks for general dimensions is not known. Likewise classification of stable policies in queueing systems is an open problem.

In modern industrial solutions, Ethernet-based communication networks have been replacing bus technologies. Ethernet is no longer found only in inter-controller or manufacturing execution systems, but has penetrated into the real-time sensitive automation process (i.e., close to the machines and sensors). Ethernet itself adds many advantages to industrial environments where digitalization also means more data-driven IT services interacting with the machines. However, in order to cater to the needs of both new and more automation-related communication, a better restructuring of the network and resources among multitenant systems needs to be carried out. Various Industrial Ethernet (IE) standards already allow some localized separation of application flows with the help of Quality of Service (QoS) mechanisms. These technologies also expect some planning or engineering of the system which takes place by estimating worst-case scenarios of possible traffic generated by all assumed applications. This approach, however, lacks the flexibility to add new services or to extend the system participants on the fly without a major redesign and reconfiguration of the whole network. Network virtualization and segmentation is used to satisfy these requirements of more support for dynamic scenarios, while keeping and protecting time-critical production traffic. Network virtualization allows slicing of the real physical network connecting a set of applications and end devices into logically separated portions or Slices. A set of resource demands and constraints is defined on a Slice or Virtual Network level. Slice links are then mapped over physical paths starting from end devices through forwarding devices that can guarantee these demands and constraints. In this chapter, the modeling of virtual industrial network constraints is addressed with a focus on communication delay. For evaluation purposes, the modeled network and mapping criteria are implemented in the Virtual Network Embedding (VNE) traffic-engineering platform ALEVIN [1].

The fluid model has proven to be one of the most effective tools for the analysis of stochastic queueing networks, specifically for the analysis of stability. It is known that stability of a fluid model implies positive (Harris) recurrence (stability) of a corresponding stochastic queueing network, and weak stability implies rate stability of a corresponding stochastic network. These results have been established both for cases of specific scheduling policies and for the class of all work conserving policies.However, only partial converse results have been established and in certain cases converse statements do not hold. In this paper we close one of the existing gaps. For the case of networks with two stations we prove that if the fluid model is not weakly stable under the class of all work conserving policies, then any corresponding queueing network is not rate stable under the class of all work conserving policies. We establish the result by building a particular work conserving scheduling policy which makes any corresponding stochastic process transient. An important corollary of our result is that the condition of the form ρ* &lt; 1, which was proven in [7) to be the exact condition for global weak stability of the fluid model, is also the exact global rate stability condition for any corresponding queueing network. Here ρ* is a certain computable parameter of the network involving virtual station and push start conditions.

The fluid model has proven to be one of the most effective tools for the analysis of stochastic queueing networks, specifically for the analysis of stability. It is known that stability of a fluid model implies positive (Harris) recurrence (stability) of a corresponding stochastic queueing network, and weak stability implies rate stability of a corresponding stochastic network. These results have been established both for cases of specific scheduling policies and for the class of all nonidling policies.\n¶\nHowever, only partial converse results have been established and in certain cases converse statements do not hold. In this paper we close one of the existing gaps. For the case of networks with two stations, we prove that if the fluid model is not weakly stable under the class of all nonidling policies, then a corresponding queueing network is not rate stable under the class of all nonidling policies. We establish the result by building a particular nonidling scheduling policy which makes the associated stochastic process transient. An important corollary of our result is that the condition ρ&lt;sup&gt;*&lt;/sup&gt;≤1, which was proven in [Oper. Res. 48 (2000) 721–744] to be the exact condition for global weak stability of the fluid model, is also the exact global rate stability condition for an associated queueing network. Here ρ&lt;sup&gt;*&lt;/sup&gt; is a certain computable parameter of the network involving virtual station and push start conditions.

We consider the problem of scheduling tasks on multiprocessor architectures in the presence of communication delays. Given a set of dependent tasks, the scheduling problem is to allocate the tasks to processors such that the pre-specified precedence constraints among the tasks are obeyed and certain cost-measures (such as the computation time) are minimized. Several cases of the scheduling problem have

Streaming media applications represent an important class of applications for embedded systems. Recent advances in design-space exploration of architectures for such applications have pointed towards the suitability of Multiprocessor System on Chip (SoC) solutions. Multiprocessor SoCs not only offer higher performance, but can also lead to solutions which are cheaper cost wise. A typical synthesis methodology for such architectures would require a validation stage at the end of final system integration. The wide availability of cheap and large FPGA devices, advances in automatic synthesis from VHDL/Verilog and abundance of high performance computing platforms enables the design of a generic validation system for such Multiprocessor SoCs.In this paper we present the design and implementation of Srijan Multiprocessor Prototyping System (SMPS). SMPS is a system for rapid prototyping and validation of single chip application specific multiprocessor systems. The individual computing elements are RISC processors, coprocessors which lie in the processor pipeline, and ASICs which connect directly to system bus. The system is a tightly coupled multiprocessor with shared memory and shared address space. A Real-time Operating System (RTOS) provides task scheduling and access to shared resources. The system is presented as a parameterized VHDL based on the open source Sparc~V8 compliant LEON processor and a homegrown light-weight RTOS, RtKer-MP. The entire VHDL is configurable using a GUI, has support for cache coherency, choice of arbitration policy and easy integration of custom processing engines. RtKer-MP allows for a pluggable scheduler, dynamic and static scheduling policies, static and dynamic task migrations domains and variable interruption frequencies for separate processors. The pluggable scheduler interface allows for quick exploration of various scheduling policies for a feedback to the estimation systems.

In this paper, we study a wireless networked control system (WNCS) with $N \ge 2$ sub-systems sharing a common wireless channel. Each sub-system consists of a plant and a controller and the control message must be delivered from the controller to the plant through the shared wireless channel. The wireless channel is unreliable due to interference and fading. As a result, a packet can be successfully delivered in a slot with a certain probability. A network scheduling policy determines how to transmit those control messages generated by such $N$ sub-systems and directly influences the transmission delay of control messages. We first consider the case that all sub-systems have the same sampling period. We characterize the stability condition of such a WNCS under the joint design of the control policy and the network scheduling policy by means of $2^N$ linear inequalities. We further simplify the stability condition into only one linear inequality for two special cases: the perfect-channel case where the wireless channel can successfully deliver a control message with certainty in each slot, and the symmetric-structure case where all sub-systems have identical system parameters. We then consider the case that different sub-systems can have different sampling periods, where we characterize a sufficient condition for stability.

Main characteristics of large-scale, geographically distributed grid systems are resource heterogeneity and network latency. In this paper, we use queuing network models to analyze data-parallel grid applications and we show the effects of resource heterogeneity, communications delays, network bandwidths and synchronization overheads on the application-level performance. The proposed models rely on the statistical pattern of computation, communication, and I/O operations in the parallel applications, as well as on measurable infrastructure characteristics. We finally show how the high variability in the execution and communication times must be considered when modeling applications on geographically distributed grid infrastructures

We consider a single server queueing system with two classes of jobs: eager jobs with small sizes that require service to begin almost immediately upon arrival, and tolerant jobs with larger sizes that can wait for service. While blocking probability is the relevant performance metric for the eager class, the tolerant class seeks to minimize its mean sojourn time. In this article, we analyse the performance of each class under dynamic scheduling policies, where the scheduling of both classes depends on the instantaneous state of the system. This analysis is carried out under a certain fluid limit, where the arrival rate and service rate of the eager class are scaled to infinity, holding the offered load constant. Our performance characterizations reveal a (dynamic) pseudo-conservation law that ties the performance of both classes to the standalone blocking probabilities associated with the scheduling policies for the eager class. Furthermore, the performance is robust to other specifics of the scheduling policies. We also characterize the Pareto frontier of the achievable region of performance vectors under the same fluid limit, and identify a (two-parameter) class of Pareto-complete scheduling policies.

The landing of a fixed-wing UAV on top of a mobile landing platform requires a cooperative control strategy, which is based on relative motion estimates. These estimates typically suffer from communication or processing time delays, which can render an otherwise stable control system unstable. Such effects must therefore be considered during the design process of the cooperative landing controller. In this letter the application of a model-free passivity-based stabilizing controller is proposed, which is based on the monitoring of energy flows in the system, and actively dissipating any given active energy by means of adaptive damping elements. In doing so, overall system passivity and consequently stability is enforced in a straightforward and easy to implement way. The proposed control system is validated in numerical simulations for round trip delays of up to 4 seconds.

A multi-Mobile Agent and optimal itinerary planning-based data aggregation in Wireless Sensor Networks