The Incredible Journey: Measuring Performance Along the Entire Stack

Abstract

During the 8 hours we spend every day in front of a computer, the greatest pain we are likely to endure comes from stress induced by application latency.

Everyone hates latency, but what are the causes? During this presentation we will take the Incredible Journey along the entire stack from user request to final response.

We will begin with time-to-first-byte, discussing DNS, CDNs, TCP and SSL handshakes.

Next we will discuss HTTP multithreading and header compression with mod SPDY, automated benchmarking to test config changes scientifically, and do a deep dive on the open source tool WebPageTest.

Then we will blow up requests within the application, looking at time spent inside the view, time in the cache, and time in the DB. We will overlay SELECTS, INSERTS, UPDATES, and DELETES in a graph and even drill down to time spent waiting for CPU vs. time spent waiting on disk.

Lastly we will take a peak at OS internals. Process stacks in userland that are working on our request, and the sys calls they make on the kernel.

When finished, attendees will know which open source tools will measure utilization and bottlenecks from front end to back. The deck will also be hosted online with links to further resources, including all code on GitHub that is used in the presentation.