Adventures in Performance

The end of the economical version of Moore's Law (exponential decrease in transistor unit costs), the advent of metered cloud computing, and the shift of the economy toward digital goods and SaaS-based business models has shifted performance engineering from the fringes ("CPU progress will fix it") to center stage: Gross margins and through that, company value is related to efficient delivery of compute-intensive services.

This talk recapitulates lessons from looking at performance and efficiency analysis of large-scale compute. Lessons include:

  • Technical: How language design choices have direct implications on performance issues
  • Historical: How the evolution of hardware leads to software that is often ill-suited for the performance geometry of the underlying machine
  • Organizational: How Google's monorepo culture vs. Amazon's two-pizza-team culture impacts code efficiency
  • Mathematical: Why statistical variance is your enemy, but really hard to control

The talk concludes with some thoughts on the inadequacy of existing tooling and where things could and should improve.


Speaker

Thomas Dullien

Distinguished Engineer, Mathematician, World-Renowned Security Researcher

Thomas started his career in the field of computer security under the pseudonym "Halvar Flake", and spent ~20 years in that field. Starting from low-level reverse engineering, he moved toward tooling for malware and vulnerability analysis, started a company that did large-scale malware disassembly and correlation, sold the company to Google, spent 7 years at Google. There, he spent time integrating & scaling the technology and then returning to a research role where he encountered Rowhammer - the DRAM-reliability-issue that he helped turn into a security issue. He then switched fields to efficient computation and cloud economics, started a company that shipped the first multi-runtime, whole-system, whole-fleet in-production profiler, and then joined Elastic as part of getting acquired. His interests are the intersection of low-level machine understanding, the economical impact of efficient compute, and the ecological benefits of greater efficiency.

Read more
Find Thomas Dullien at:

Date

Wednesday Mar 29 / 10:35AM BST ( 50 minutes )

Location

Whittle (3rd Fl.)

Topics

performance cloud monorepo

Share

From the same track

Session performance

Performance: Adventures in Thread-per-Core Async with Redpanda and Seastar

Wednesday Mar 29 / 11:50AM BST

Thread-per-core programming models are well known in software domains where latency is important.

Speaker image - John Spray

John Spray

Storage Engineering Lead @neon.tech, Formerly Redpanda, Inktank (Ceph), Whamcloud (Lustre)

Session performance

What We Talk About When We Talk About Networks

Wednesday Mar 29 / 02:55PM BST

Networks, and the applications they support, sometimes treat each other as strangers.

Speaker image - Keith Winstein

Keith Winstein

Associate Professor of Computer Science and, by courtesy, of Electrical Engineering @Stanford

Session performance

Providing a Personalized Experience to Millions of Users @BBC

Wednesday Mar 29 / 01:40PM BST

Personalization can be a very effective and impactful way for your organisation to attract new customers as well as to retain your existing customer base – if you get it right!

Speaker image - Manisha Lopes

Manisha Lopes

Principal Software Engineer @BBC

Session performance

WebGPU is Not Just About the Web

Wednesday Mar 29 / 04:10PM BST

Since OpenGL has been deprecated, writing portable native apps that leverage the massively parallel computing capabilities of GPUs has become tedious because no single graphics API can fit all targets.

Speaker image - Élie Michel

Élie Michel

Research Scientist @Adobe