Download as PDF, PPTX
Uploaded byl xf
PDF, PPTX3,578 views
Scalable Networking
The document discusses how to write scalable network servers that can handle high loads with many concurrent connections. It begins by providing some background on the author and why high performance network code is important. It then shows simple examples of a basic HTTP client and server implementation. The document discusses issues with the traditional approach of using one process per connection, including the high latency of forking processes and scheduling overhead with many processes. It presents benchmark results demonstrating the performance costs of forking. The document advocates avoiding the one process per connection model for scalability.
In this document
Powered by AI
Introduction to scalable network programming, need for high-performance servers to handle thousands of clients, examples of server configurations under high load.
Basic implementation of a web client and server, discussion of necessary elements like error handling and scalability.
Challenges of using a one-process-per-connection model, and performance metrics related to forking processes.
Importance of scheduling in OS for managing multiple processes, Linux scheduler improvements and structure.
Fork performance implications, issues with memory consumption, and implications of copy-on-write during forks.
One-process-per-connection model advantages, shortcomings of traditional models, analysis of performance metrics.
Comparison of threads and processes for scalability in network programming, introduction of thread pools.
Methods for timeout handling, including Unix signals, select, and poll functions for managing active connections.
Event-driven programming models like epoll in Linux, performance optimizations for handling large connections.
Epoll use case scenarios, integration into network programming, and comparisons with FreeBSD's kqueue.
Impact of file descriptors, fragmentation, and memory management on network latency, analysis of data rates.
Limitations of asynchronous I/O in current systems, its implementation challenges across different operating systems.
Recommendations for improving server performance, comparison of latency across different versions of network OS.
Wrap-up of presentation, resources for further information, and contact details for the presenter.