An approach to identify performance influence on web services, using different parallel execution techniques

Abstract

Communication between software services become an essential requirement in modern software industry. With the improvement of internet services, connecting existing software services with each other and provide unlimited services via the internet to the end users become a crucial requirement for modern businesses. Web services have been introduced as the standard communication medium between software services and micro services-based architecture which enrich the usage of web service. IoT services and increasement of smart devices create high demand for web services. Due to those reasons, performance of web services become very critical and crucial to the existing and new software services. Web service performance can improve by distributing multiple instances among multiple nodes where the approach is not always cost effective. Other possible approach would be parallelly executes the web service requests. Multi-core architecture has been introduced to execute computer programs parallelly in processor level. To get optimum performance from this multi-core processors, computer programs should be able to execute parallelly at processor cores level. Multi-threaded and Multi-core based parallel execution techniques were mainly evaluated by implementing prototype REST APIs for matrix multiplication. REST API performance impact was evaluated against the processor affinity. When processing complexity increases in serial processing REST API, their performance (measured by throughput) decreases rapidly. When complexity increases for REST end points where they have parallelly execution methods, their performance also decreases but decrease amount is smaller by comparing to the serial processing. There is notable performance gain for multi-process parallelly executions over the multi-threaded parallel execution. Best throughput was achieved, when multi-core process binds to specific core of the processor (applying processor affinity).