Enhancing the scalability of SPDY clients for testing with limited network resources

Currently there are many tools available to generate HTTP loads to test and benchmark the performance of hosted web services such as Amazon Web Service. However, none of these tools currently supports SPDY protocol with a number of concurrent client requests. SPDY over HTTP is recently becoming more popular for web service providers, hence it is desirable to develop tools to benchmark the performance of SPDY enabled servers or clients. In this paper we describe methods by which we can use SPDY network resources optimally for testing purposes, producing more number of network requests with less number of clients. We demonstrate the horizontal scalability of currently supported SPDY network analysis tools and how to overcome their system limitations such as the 64K limit on connections. We also study the tradeoff between the number of threads and the number of processes while generating network loads for testing. Our enhancements include modifications to the spdycat tool to generate the loads for a higher number of users. Our experimental setup consists of cluster of ATS instances hosted on AWS on the backend of the load balancer, and successfully simulate one million active users with two requests per user.