Go fast

Reference Materials > https://github.com/Randommood/go-fast

We all want to move fast. We construct our systems in a rapidly-iterating and agile way. We design and build them to be efficient, robust, and have low latency. But sometimes in the search for speed we make decisions that come back to haunt us. Some of these are reasonable tradeoffs, some can have long-term repercussions, and some can be outright catastrophic.

At Fastly, we serve over 10% of the internet’s traffic and power tens of thousands of online destinations. We have a unique vantage point on the state of the web and going slow is not an option. When Beyonce tweets a baby photo, the NYT breaks a story, or taylor swift drops a new album on spotify … our edge cloud sees it first and brings it to you.

In this talk we’ll go over the key lessons we learned while building our edge cloud services: what worked, what didn’t, the tradeoffs we made, and what can you learn from our experiences while building your own distributed applications & teams at scale.

Transcript

1. What does it take to go FAST
2. Hi, I’m Inés Sombra @Randommood Thank you! I’m excited to be here
3. WHY W S SYSTEMS T TAKEAWAYS Today’s Agenda
4. 2013?
5. Artur Bergman
6. Why go
 Fast?
10. Edge Cloud Services
11. Edge Cloud Service Constraints Highly Available Globally distributed Horizontally Scalable Secure Self healing Gracefully degrade under faults* Latency sensitive 
 and fast! (* when appropriate)
12. Your Constraints Efficient computations Elastic platform Secure Fast & responsive ($$$)
13. Lara Hogan 2017
14. Optimizing the images in your site is a great way to make it fast Look for optimizations you can make based on image format, quality, & display needs Image pre-processing is a way to do this Lara’s Lessons ! " # ∠ ∠ ∠ ∠ ∠ ' www.fastly.com
15. JPEG 1920 × 1080 px Max quality 2.3MB Original image
16. Hero image Thumbnail image 200 × 200 px Quality 85 6.93KB vs 2.3MB JPEG 800 × 560 px Quality 85 47 KB vs 2.3MB
17. Edge Cloud Services
18. Image Optimization Service
19. Building 
 Systems (
20. Building New Systems Fast Leverage orthogonality Invest in operational excellence Aggressively narrow down scope and set a MVS* (* Minimum Viable Speed)
21. Orthogonality / Composability ) + + + System Goal Constraints Small Team
22. Orthogonality / Composability CDN for request management CDN for logging CDN for authentication CDN for state management CDN for purging CDN for API translation CDN for fault tolerance CDN for doing less work!
23. Image Optimizers Origin
24. gordo-thumb please! Image Optimizers Origin
25. gordo-thumb please! original gordo please! Image Optimizers Origin
26. gordo-thumb please! there you go Image Optimizers Origin
27. gordo-thumb please! gordo-thumb please! Image Optimizers Origin
28. Image Optimizers gordo-thumb please! Done! Origin
29. gordo-thumb please! there you go Image Optimizers Origin
30. YASS! Image Optimizers Origin
31. gordo-thumb please! Image Optimizers Origin
32. YASS! Image Optimizers Origin
33. Origin Orthogonality & You Image 
 Pre-Processing
34. Origin Orthogonality & You Image 
 Optimization Reduce storage cost in origin
35. Orthogonality & You Your System Origin Image 
 Optimization
36. Doing less is awesome, look for ways in which you can leverage composability Moving logic & data closer to end users increases speed & efficiency (if you can) Your system is everything from your end users to your origin Leverage Orthogonality
37. ImageOpto POP Invest in Operational Excellence
38. ImageOpto GCP ImageOpto POP Invest in Operational Excellence
39. Image Optimizers Invest in Operational Excellence
40. Image Optimizers Invest in Operational Excellence
41. Image Optimizers Invest in Operational Excellence
42. Image Optimizers Invest in Operational Excellence
43. Image Optimizers Invest in Operational Excellence
44. Image Optimizers Invest in Operational Excellence
45. Complex operations make systems less resilient, more incident-prone, and more difficult to manage Pay-as-you-go investment model in system operability Pay attention to deployment cadence & other metrics* Invest in Operational Excellence (* PRs open & age, number of incidents)
46. Aggressively Narrow down your scope
47. Aggressively Narrow down your scope Decodes from an uncompressed format are faster About 90% reduction in processing time by saving the decoded image into an uncompressed format Let’s use this strategy to provide speedier transformations! *
48. Aggressively Narrow down your scope Design resulted in a more complex request cycle. What happened in this request? was hard to answer Starting with a non-trivial solution took a lot of time Unexpected interactions with purging & shielding !
49. Set your Minimum Viable Speed Setting an SLA is hard when you are creating a new service The network can make you slow, things you didn’t know you had (or that you were not doing well) can make you slow Pick a MVS to get started, optimize & learn as you go, then formalize your SLA
50. Fast enough is good enough* Discipline yourself to ship something first & expand your API as needed Tracking and fixing “its slow” is an ongoing activity Narrow Scope & Setting your MVS (* super context specific)
51. Building trust is key Focus on the whole team’s output Metrics to watch: PRs open & age, deployment cadence, review speed, etc Strive to create a sense of ownership 
 & pair it with the ability to have impact Supported By your Organization
52. Camille 
 Fournier
53. ( Some
 Takeaways
54. Tradeoffs are made in context and should be revisited often Goal tunnel vision may lead you to work harder on the wrong solution Narrowing focus first and expand it as you learn is a solid approach All about the Tradeoffs
55. The less to do the faster you can go, reduce your scope! Keep the global system context in mind Move as much logic & data to the edge as you can for speed & efficiency SYSTEMS TL; DR; ) + ORGS Building trust is key Focus on the whole team’s output Invest in operability There are many paths to an acceptable solution (watch out for NIH)
56. https://github.com/Randommood/go-fast @Randommood Special thanks to: James Turnbull, Tyler McMullen, Joao Taveira, Elaine Greenberg, Peter Teichman, Lorenzo Saino, Adam Thomason, & Greg Bako Remember to optimize your images! THANK YOU