Comparing the performance of web server architectures

Abstract

In this paper, we extensively tune and then compare the performance of web servers based on three different server architectures. The µserver utilizes an event-driven architecture, Knot uses the highly-efficient Capriccio thread library to implement a thread-perconnection model, and WatPipe uses a hybrid of events and threads to implement a pipeline-based server that is similar in spirit to a staged event-driven architecture (SEDA) server like Haboob.We describe modifications made to the Capriccio thread library to use Linux's zero-copy sendfile interface. We then introduce the SYmmetric Multi-Processor Event Driven (SYMPED) architecture in which relatively minor modifications are made to a single process event-driven (SPED) server (the µserver) to allow it to continue processing requests in the presence of blocking due to disk accesses. Finally, we describe our C++ implementation of WatPipe, which although utilizing a pipeline-based architecture, excludes the dynamic controls over event queues and thread pools used in SEDA. When comparing the performance of these three server architectures on the workload used in our study, we arrive at different conclusions than previous studies. In spite of recent improvements to threading libraries and our further improvements to Capriccio and Knot, both the event-based µserver and pipeline-based WatPipe server provide better throughput (by about 18%). We also observe that when using blocking sockets to send data to clients, the performance obtained with some architectures is quite good and in one case is noticeably better than when using non-blocking sockets.

Hop, a Fast Server for the Diffuse Web

“…In addition to be fast because it avoids data copies and user-space/kernel-space traversals, it also simplifies the implementation of servers because it makes memory caches useless for serving static files. As in a previous study [29], we have noticed no acceleration when using memory cache for serving files instead of using a pure sendfile-based implementation. Using sendfile or a similar function actually delegates the caching to the operating system which is likely to perform more efficiently than a user-land application.…”

“…Initially it has been observed that kernel-space servers outperformed user-space servers [13], independently of the hosting operating systems. Adding zero-copy functions in the system API, such as the sendfile, has changed the conclusion [29]. In addition, another study [35] has shown that the gap between kernel-space and user-space that prevailed in older implementations unsurprisingly becomes less significant when the percentage of requests concerning dynamically generated content increases.…”

“…This independence also allows many Web architectures to be used for scheduling the Hop pipeline. For instance, Hop may use multi-processes, multithreads, pipeline [44,9], AMPED [28] or SYMPED [29], or a mix of all of them. In order to avoid early decisions, we have extracted the Hop scheduler from the core implementation of the server.…”

“…Non-copying output Several studies have measured the benefit to be expected from using non-copying output functions such as the Linux sendfile facility [26,29]. This system service is supported by Hop.…”

X-Switch: An Efficient , Multi-User, Multi-Language Web Application Server

“…The X-Switch system uses a polling mechanism to support multiple connections. Studies by Pariag et al [13] revealed that an event-driven or hybrid server achieved up to 18% higher throughput than the best implementation of the thread based server. Therefore the X-Switch system uses a single thread to listen to and poll all its connections.…”