September 19–20, 2016: Training
September 20–22, 2016: Tutorials & Conference
New York, NY

High-performance servers: Lessons from the real world

Samy Al Bahra (Backtrace I/O)
4:45pm–5:25pm Wednesday, 09/21/2016
Infrastructure reimagined Cloud, Networking Beekman Audience level: Intermediate
Average rating: ****.
(4.33, 3 ratings)

Prerequisite knowledge

  • A basic understanding of networking and Unix systems
  • What you'll learn

  • Learn patterns and anti-patterns to high-performance server design and implementation, as well as techniques to diagnose performance problems
  • Description

    In this world of horizontal scalability in the cloud, the art and science of high-performance server architecture has been neglected and relegated to the iron curtain of proprietary software. The right design and implementation choices dictate your user experience and bottom line. Samy Al Bahra examines key insights from research and practice in high-performance server architecture, drawing on examples from the servers that drive a majority of today’s online messaging and advertising.

    There is a theme to the scalability bottlenecks that forces servers into longer-tail performance degradation. Whether your workload is CPU-bound, I/O-bound, or a mix of both, techniques have been developed over the last two decades that allow you to identify and mitigate a lot of pain points, and design strategies have been developed to match your workload. Samy introduces key principles in computer architecture, operating systems, concurrency, storage devices, and quality of service that allow you to better approach server design and implementation and addresses topics ranging from multicore synchronization to cache pollution and I/O bottlenecks. Samy begins with the seminal research in server design, including a tour of common server architecture idioms by workload, and along the way injects real-world examples and common mistakes (and their side-effects) on the path to implementation and supporting nonuniform workloads.

    Photo of Samy Al Bahra

    Samy Al Bahra

    Backtrace I/O

    Samy Al Bahra is the cofounder of Backtrace, where he is helping build a modern debugging platform for today’s complex applications. Prior to Backtrace, Samy was a principal engineer at AppNexus, where he played a lead role in the architecture and development of many mission-critical components of the ecosystem. His work at AppNexus was instrumental in scaling the system to 18 billion impressions with orders of magnitude in efficiency improvements. Prior to AppNexus, Samy was behind major performance improvements to the core technology at Message Systems. At the George Washington University High Performance Computing Laboratory, Samy worked on the UPC programming language, heterogeneous computing, and multicore synchronization. Samy is also the founder of the Concurrency Kit project, which several leading technology companies rely on for scalability and performance. Samy serves on the ACM Practitioners Board.