Benchmarking the request throughput of conventional API calls and gRPC : A Comparative Study of REST and gRPC

Abstract: As the demand for better and faster applications increase every year, so does the demand for new communication systems between computers. Today, a common method for computers and software systems to exchange information is the use of REST APIs, but there are cases where more efficient solutions are needed. In such cases, RPC can provide a solution. There are many RPC libraries to choose from, but gRPC is the most widely used today. gRPC is said to offer faster and more efficient communication than conventional web-based API calls. The problem investigated in this thesis is that there are few available resources demonstrating how this performance difference translates into request throughput on a server. The purpose of the study is to benchmark the difference in request throughput for conventional API calls (REST) and gRPC. This was done with the goal of providing a basis for making better decisions regarding the choice of communication infrastructure between applications. A qualitative research method with support of quantitative data was used to evaluate the results. REST and gRPC servers were implemented in three programming languages. A benchmarking client was implemented in order to benchmark the servers and measure request throughput. The benchmarks were conducted on a local network between two hosts. The results indicate that gRPC performs better than REST for larger message payloads in terms of request throughput. REST initially outperforms gRPC for small payloads but falls behind as the payload size increases. The result can be beneficial for software developers and other stakeholders who strive to make informed decisions regarding communication infrastructure when developing and maintaining applications at scale.