Answer

Abstract

Distributed systems are a crucial area in computing that has attracted a lot of research. Most often, distributed systems experience optimization challenges because of shared resources. This paper recommends the use of mathematical models that are applied in the allocation of algorithms to balance the usage of resources in a distributed computing environment. It applies stochastic reward nets to ensure that algorithms are used ergonomically. This calls for extensive research since distributed systems experience optimization challenges as most resources are shared. Researchers have embarked on ways of analyzing the best ways of acquiring the load information of nodes to improve performance. Furthermore, they have developed powerful models that can isolate the most appropriate migration methodologies in different computing environments. Besides, this paper also addresses the challenges by proposing the most viable solutions. A power-aware strategy prioritizes power consumption to increase the processing rate. The paper recommends the use of these two strategies and suggests the best scenarios where the strategies can apply. Moreover, selecting the most appropriate migration process has a direct impact on the performance of the system. The paper concludes by suggesting the best strategy to be utilized in the process of allocating algorithms to processes in a distributed system environment.      

Allocation of Algorithms and Process Migration

Some of the major concerns in a distributed computing environment are incurring low cost while at the same time maintaining high performance of the system (Asadi et al., 2020). These concerns have prompted organizations, scientific firms, and defense organizations to come up with the most sufficient algorithms. Well written algorithms are aimed at solving load balancing issues in any given distributed computing environment (Asadi et al., 2020). These algorithms assist in resource allocation in multicore systems using   Non-Uniform Memory Access(NUMA)architecture(Chiang et al.,2018). However, distributed computing systems experience many drawbacks such as low performance(Chiang et al.,2018). This results from remote memory access and resource allocation in the Non-Uniform Memory Access systems(Chiang et al.,2018). In the assessment of power consumption, some of the hardware that is considered are central processing units, network interfaces, and storage. This paper comprehensively reviews the best power migration strategies suitable in a given distributed system environment. Mathematical models such as stochastic reward nets methodology are used to achieve this.

Distributed systems utilize pool computing resources to achieve a significant level of computational power (Asadi et al., 2020). The system management possesses comprehensive data about all the workstations and processes running at any particular time. The users in a distributed system environment utilize the system’s resources at any given time. This raises the issue of complexity (Asadi et al.,2020). Nearly all distributed systems utilize multi-processor architecture. Sometimes, the load distribution is decreased due to the data about the whole processes as well as the existing resources prompting the load balancer to relocate some processes. Through displacing some of the processes, the load balancer can reload balancing. An increase in the need for more computing power presents the issue of load balancing. There is static load balancing where a local task can be transferred to a remote node or can be designated to a threshold queue from an assignment queue. On the other hand, in dynamic load balancing, computational jobs are reallocated at runtime (Asadi et al.,2020). This means that the load moves from heavily loaded nodes to nodes with lighter tasks. The major aim of load balancing is improving the efficiency of a distributed system by ensuring that the application load is distributed evenly. The load balancer runs various algorithms to determine how to proceed with process migration. When a process transmission has occurred successfully, another load balancing pattern is created.

Partial or whole of running processes from one node to another is done through process migration. This kind of migration enables a system to easily run on any node within a distributed system environment. In a cluster computing system, choosing a good process migration technique will have a positive impact on the system’s functionality. In case the process migration is not appropriate for a distributed computing environment asset then process migration time surges exponentially (Asadi et al.,2020). However, in a cluster computing system, only one task can be executed at any given time. Processes are generated in a cluster computing environment to be handled by the computing nodes. The processes can communicate with each other to utilize their dependencies thus preventing any resource conflicts with other processes executing in the local machine (Asadi et al.,2020). At run-time, the computing nodes cannot be changed. In this paper, mathematical models are applied by the use of regression in the calculation and prediction of the total power consumption.

Motivation for Research

Implementing a distributed computing system in an organization comes with many benefits. A major advantage of a distributed computing system is the improvement in performance and computing power in a system. Distributed computing systems provide more resources such as processors to speed up the execution of tasks. Furthermore, the application of these systems minimizes costs while increasing functionality in the entire system architecture (Van Steen, Maarten & Tanenbaum, Andrew, 2016). Moreover, resources can easily be shared in a distributed computing environment. These resources include both hardware and software components. Users can access pooled resources as long as they are connected to the distributed computing environment (Van Steen, Maarten & Tanenbaum, Andrew, 2016). Besides, defective hardware can easily be replaced in a distributed computing environment without affecting others connected to the system. Workstations can still operate after the system that failed is removed (Van Steen, Maarten & Tanenbaum, Andrew, 2016). Furthermore, increasing the number of computers in a distributed computing environment is possible making it scalable (Van Steen, Maarten & Tanenbaum, Andrew, 2016). Besides, power wastage in a distributed system is minimized as power is consumed simultaneously by all the hardware components and applications in the system.

Literature Review

Long before the advent of distributed computing companies, scientific research laboratories, military and defense agencies, and healthcare industry relied on single process computing. With time multitasking operating systems were adopted for use in multinational companies. Many tasks could run at any given time. However, this was not sufficient enough as there was a need for computers to handle more complex tasks (Fabra,2019). This need led to the introduction of client-server architecture. Client-server architecture applied remote procedure calls (RPC) (Fabra,2019). Remote procedure calls permitted a program running on a particular system to be able to communicate with another program running on another system. With the continuous development of powerful networks, virtual machines in a network could communicate with each other. Objects present in the virtual machines could now share information as long as they were connected to the same network (Fabra,2019). The functionalities of the virtual machines formed the basis of distributed computing. Computers now began communicating through socket-based remote connectivity. This technology provided a two-way communication link between programs running on the same network. Socket based remote connectivity supported multi-threading. With advancements in technology, there was a need to improve from a common client-server to a multi-node distributed system. Workstations could now be used as both clients and servers (Fabra,2019). With powerful networks connecting different organizations and institutions, personal computers could now communicate with each other and share resources such as printers. This made organizational computers operate as servers as well as clients. This created a fully functional distributed system (Fabra,2019). There was no stopping and since then, very powerful distributed systems exist almost everywhere in the world.

Problem Statement

The need for powerful distributed computing systems such as an exascale is inevitable for any organization in the world. The fast computing systems come with several merits such as high performance that significantly enhances the system’s efficiency. For instance, scalability is a crucial area in computing that must be taken into consideration (Asadi et al., 2020). The system should be scalable to ensure that other components can be added when the need arises. A scalable system will allow an individual to easily get data of any machine in the system. Furthermore, heterogeneity is a core issue that must be addressed during the implementation of distributed computing systems (Asadi et al., 2020). Heterogeneity allows the integration of various processes and ensures that shared resources such as processors and shared memory are accessible by all the connected workstations (Asadi et al., 2020). Also, it significant to find ways of receiving and processing the load information of the entire system. Therefore, improving the system’s effectiveness requires a methodology that will seamlessly adapt to all the processes.Stochastic reward nets choose the best strategies that can be applied in process migration (Asadi et al., 2020). The mechanism gives a strong modeling environment because of its performance, analysis of performability as well as its dependability (Asadi et al., 2020). The stochastic reward nets use two models that expound on resource allocation and process migration respectively. Moreover, reading the chosen process on the host source is a bottleneck because of synchronization issues. Besides, the rebuilding of the migration process in the destination host is affected by performance and scalability issues.

Methodologies

Some of the models used in this paper encompass stochastic reward nets. This mathematical model studies how power consumption and performance are affected by the algorithms of allocating resources and process migration. This model assists in assessing a system’s behavior. According to Asadi et al. (2020), stochastic reward nets consist of two elements known as places and transitions. Places are holders of tokens and circular in nature. Tokens are the dots inside places differentiating information moving from one point to another. On the other hand, transitions are the paths that are used to move tokens from one place to another place (Asadi et al., 2020). They are rectangular in nature. Stochastic reward nets methodology is a modeling system that allows seamless analysis of complex systems in terms of performance and reliability. However, a major drawback experienced when using stochastic reward nets is when there is an explosion of state. This arises when there is an upsurge of components in the model. However, various assumptions are considered when working with this model (Asadi et al., 2020). These assumptions will affect the final results making them deviate a little in an existing distributed computing system. To make this model functional, some figures can be approximated so as not to deviate far from those in the real-world distributed computing system. Some of these approximation methods include lumping quotient and folding techniques.

Results and Findings

From the Stochastic reward nets methodology, both performance and computing power and performance are calculated concerning a real-world distributed computing system (Asadi et al.,2020). Key factors that are put into consideration are blocking probability of processes, use of computers, throughput, and power consumption.

Obtaining load information of the machine.

 To obtain load information, the process follows a specific queue (Asadi et al.,2020). Once the queue is full, any process that arrives is immediately blocked. The probability that a process will be blocked depends on the number of tokens present in the places. Besides, according to the Stochastic reward nets model, a workstation is active only if there is a task assigned to it (Asadi et al.,2020). This is calculated by equating the total number of tokens in the places. If the number of tokens in the chosen computer does not result in zero, then it is termed as active. On the other hand, if the number of tokens in it equates to zero, then the selected computer is in an inactive state.

Communicating with each extension of the system

A set threshold cannot be surpassed for processes running on a distributed computing environment. Whenever the number of processes surpasses the set target, processing proportion significantly reduces. Moreover, from the model, it is noted that only work stations that have running processes are active. On the other hand, inactive workstations have a power consumption rate of zero. This can be expressed in an equation as:

Each extension of the system should communicate very well with each other to support process migration(Asadi et al., 2020).

Numerical Results

Receive and process the load information of the whole distributed system

From the Stochastic reward nets methodology, the rate of arrival of processes is the total of arrival rates of compute-, data-, and communication-intensive processes. It is mathematically proven that an increase in arrival rates is directly proportional to an increase in, blocking probability, power consumption, and throughput when using diverse resource allocation algorithms.

Process Migration selection strategies

The process migration method was conducted through power-aware and performance-aware strategies. When using a performance aware migration strategy, there is a great improvement in the blocking probability of processes. This is because there is an even distribution of processes performance-aware that raises the rate of processing in a distributed computing environment(Asadi et al., 2020). However, when a power-aware strategy is used, the blocking probability of processes reduces. Power-aware strategy only used when power consumption is the only consideration taken(Asadi et al., 2020). However, it is inversely proportional to other factors such as throughput and blocking probability. It means that tasks or processes will be migrated to workstations with more power. Moreover, in power-aware migration, the usage of power improves while the throughput rate decreases.

Reading the selected process on the host computer

Performance-aware migration requires that if a computer contains the maximum number of processes that number surpasses a set target it is selected as the source computer(Asadi et al., 2020).On the other, in power-aware migration, the host computer should have the lowest number of processes that are to be migrated.

Rebuilding the selected process on the destination  computer

In a performance-aware migration strategy, the number of processes should not surpass the threshold after process migration is achieved(Asadi et al., 2020). On the other hand, in a power-aware migration strategy, the number of processes should be at the minimum value.

Related Work

There have been several works done by other researchers in the process of migration in a distributed system environment. For instance, an MPDM mechanism is an approach used by other researchers (Khaneghah et al., 2018). In this approach, five parameters are applied. In the MPDM mechanism, the process information is stored in the process of migration management. Besides, other exceptional research done by these experts can be categorized into classes such as class I, class II and class III. In class, I category, the performance of the computing system is given preference at the expense of the arising demerits (Khaneghah et al., 2018). In class I, multi-strategies are applied for process migration. Besides, Class II systems are designed and optimized in such a way that it applies Flushing, Post Copy, and Pre-Copy techniques in its operations. These methods permit process migration to and lower overheads such as costs of hardware. Systems in this class operate on three stages. During the initial stage known as the pre-migration stage, the processes are sent to the file server. In the second phase, the process and its current state are sent to the destination server (Khaneghah et al., 2018). In the final stage, the private data is sent to the destination computer while the other part of it directed to the file server (Khaneghah et al., 2018). Class III systems are made in such a way that they utilize the blocking probability, power consumption, throughput as and utilization of each work station. An example relevant to this class is Pre-Record Algorithm that operates based on dependencies and faults in a system. The variation in the execution of processes in the host and destination computer is computed by this algorithm.

Comparison and Future Work

The use of Stochastic reward nets methodology closely relates to the work that has already been done by researchers in the field of distributed computing systems. For instance, Class III systems apply the four basic strategies in their operations (Khaneghah et al., 2018). The strategies include block probability, power consumption, throughput as and utilization of each work station (Khaneghah et al., 2018). The techniques enhance the scalability of computers in the distributed computing environment significantly (Khaneghah et al., 2018). Future researchers will have a great foundation since the strategies applied in most computers in a distributed environment are nearly the same.

Conclusion

Proper load balancing in a distributed system enhances the operations of workstations significantly. Besides, well-written algorithms reduce the execution times of processes in a distributed computing environment. Moreover, process migration needs to follow articulate strategies to be achieved successfully. The drawbacks experienced in the Non-Uniform Memory Access multicore systems can be solved by using the appropriate process migration strategies. Some of the techniques put into consideration during process migration include block probability, power consumption, throughput as and utilization of each work station. However, these techniques are normally affected depending on the strategy applied. For instance, if one selects a power-aware strategy, the rate of power consumption improves whereas throughput reduces. On the other hand, when performance aware strategy is picked, block probability improves whereas power consumptions decrease. Moreover, the processing rate is enhanced hence increasing the performance of a work station. In most legacy systems, Flushing, Post_Copy, and Pre_Copy are used for the optimization of the process migration. The legacy systems are classified into Class I, II, and III depending on their various functionalities. However, current systems borrow a lot from Class II systems in terms of functionality and strategies applied. For instance, the basic strategies used in Class III systems are the ones applied in the current mathematical models such as Stochastic reward nets methodology.