A Comparative Performance Analysis of Load Balancing Algorithms in Distributed System using Qualitative Parameters

Abstract

Static load balancing policies are generally based on the information about the average behavior of system; transfer decisions are independent of the actual current system state. Static load balancing schemes use a priori knowledge of the applications and statistical information about the system. In static load balancing, the performance of the processors is determined at the beginning of execution. Then depending upon their performance the work load is assigned by the master processor. The slave processors calculate their allocated work and submit their result to the master. A task is always executed on the processor to which it is assigned that is static load balancing methods are non preemptive. The goal of static load balancing method is to reduce the execution time, minimizing the communication delays. A general disadvantage of static approaches is that the final selection of a host for process allocation is made when the process is created and cannot be changed during process execution to make changes in the system load. There are four types of static load balancing:-Round Robin algorithm, Randomized algorithm, Central Manager Algorithm, and Threshold algorithm. 3.1 Round Robin Algorithm Round Robin algorithm [5] distributes jobs evenly to all slave processors. All jobs are assigned to slave processors based on Round Robin order, meaning that processor choosing is performed in series and will be back to the first processor if the last processor has been reached. Processors choosing are performed locally on each processor, independent of allocations of other processors. The main advantage of Round Robin algorithm is that it does not require inter process communication. In general Round Robin is not expected to achieve good performance in general case. However when the jobs are of unequal processing time this algorithm suffers as the some nodes can become severely loaded while others remain idle. Round Robin is generally used in web servers where generally HTTP requests are of similar nature and thereby be distributed equally. 3.2 Randomized Algorithm Randomized algorithm [5] uses random numbers to choose slave processors. The slave processors are chosen randomly following random numbers generated based on a statistic distribution. Randomized algorithm can attain the best performance among all load balancing algorithms for particular special purpose applications. 3.3. Central Manager Algorithm Central Manager Algorithm [6], in each step, central processor will choose a slave processor to be assigned a job. The chosen slave processor is the processor having the least load. The central processor is able to gather all slave processors load information, thereof the choosing based on this algorithm are possible to be performed. The load manager makes load balancing decisions based on the system load information, allowing the best decision when of the process created. High degree of inter-process communication could make the bottleneck state. This algorithm is expected to perform better than the parallel applications, especially when dynamic activities are created by different hosts. 3.4. Threshold Algorithm In Threshold algorithm [6], the processes are assigned immediately upon creation to hosts. Hosts for new processes are selected locally without sending remote messages. Each processor keeps a private copy of the system's load. The load of a processor can characterize by one of the three levels: under loaded, medium and overloaded. Two threshold parameters t_under and t_upper can be used to describe these levels. Under loaded: load < t_under , Medium : t_under ≤ load ≤ t_upper , Overloaded: load > t_upper. Initially, all the processors are considered to be under loaded. When the load state of a processor exceeds a load level limit, then it sends messages regarding the new load state to all remote processors, regularly updating them as to the actual load state of the entire system. If the local state is not overloaded then the process is allocated locally. Otherwise, a remote under loaded processor is selected, and if no such host exists, the process is also allocated locally. Thresholds algorithm have low inter process communication and a large number of local process allocations. The later decreases the overhead of remote process allocations and the overhead of remote memory accesses, which leads to improvement in performance. A disadvantage of the algorithm is that all processes are allocated locally when all remote processors are overloaded. A load on one overloaded processor can be much higher than another overloaded processor, causing significant disturbance in load balancing, and increasing the execution time of an application [5]. IV. DYNAMIC LOAD BALANCING ALGORITHMS In dynamic load balancing algorithms work load is distributed among the processors at runtime. The master assigns new processes to the slaves based on the new information collected [7]. In a distributed system, dynamic load balancing can be done in two different ways: distributed and non-distributed. In the distributed one, the dynamic load balancing algorithm is executed by all nodes present in the system and the task of load balancing is shared among them. The interaction among nodes to achieve load balancing can take two forms: cooperative and non-cooperative [8]. In cooperative, the nodes work side-by-side to achieve a common objective, for example, to improve the overall response time, etc. In non-cooperative, each node works independently toward a goal local to it, for example, to improve the response time of a local task. Dynamic load balancing algorithms usually generate more messages than the non-distributed ones because, each of the nodes in the system needs to communicate with every other node. A advantage of this is that even if one or more nodes in the system fail, it will not cause the total load balancing process to halt, it instead would affect the system performance to some extent. In non-distributed type, either one node or a group of nodes do the task of load balancing. Non-distributed dynamic load balancing algorithms can take two forms: centralized and semi-distributed.