P99 Conf 2023: High Performance, Low Budget

It is one thing to solve performance challenges when you have plenty of time, money, and expertise available. Many performance experts have deep knowledge of a specific system and work on large-scale problems where investing time and effort to get a fantastic result is the norm.

But what do you do in the opposite situation? If you are a small startup with no time or money and still need to deliver outstanding performance? When you may have experts in parts of the stack, but performance has to be delivered across the entire application and infrastructure stack?

It may be tempting to think that as a small startup, you can just ignore performance until you scale - but your customers will soon tell you otherwise. In this talk, Gwen talks about her journey from in-depth Kafka performance to leading a small team of engineers to ship a high-performance MVP. She'll share the decisions that worked well, the mistakes made, and the tools that were worth far beyond their cost.

Transcript

1. HOSTED BY High Performance, Low Budget Gwen Shapira Co-founder @ Nile
2. Gwen Shapira Co-founder @ Nile ▪ I used to lead the Cloud Native Kafka org at Confluent ▪ Wrote books, gave talks, have a podcast ▪ I am obsessed with performance ▪ Mostly of data systems and below ▪ but I’m also leading a scrappy startup
3. This talk is “sample size = 1” I share stories, lessons, tradeoffs and decisions from my experience in the last 2 years.
4. “Don’t worry about performance”
5. Performance expectations should be part of the early product discovery process.
6. Is Performance a core value of your product?
7. Why not just “ship and see?”
8. Survivor Bias You only get feedback from non-representative sample of users.
9. Hyper-focus on User Experience
10. User Experience of Performance
11. Web vitals are all about user’s perceived latency.
12. Login page - Serverless compute - Great dev ex, but… - Your instance will be idle. A lot. - CDNs - Optimize app start times - Provisioned compute
13. Dashboard 1. Parallelized requests 2. API improvements 3. Cache results in browser 4. DB tuning 5. Moved the page🥇
14. Effective Benchmarking
15. Benchmarks should “pay rent” by answering a question
16. Is our DB setup good? - Typically uses a standard benchmark like TPCC - Increase load over time - Look for bottlenecks - Fix what you can
17. Well? - Running a standard benchmark takes time and effort. - We fixed some configuration issues. - We had some idea of what performance to expect. Score: ⭐⭐
18. Do we deliver good experience?
19. What is a good user experience? - User On-boarding - Anything annoyingly slow - Anything your target customers obsess over
20. Customers obsess over things
21. Keep tests stupid simple
22. Simple test tips that helped us - Include extra stupid sanity tests - Check return values - Actually analyze the results Score: ⭐⭐⭐⭐
23. Benchmark shows a request that takes 15s on avg. Why?
24. Prioritize fixes based on expected ROI. Parsing Lookups Real work Request Time
25. Have no time to optimize? At least don’t pessimize.
26. @gwenshap gwen@thenile.dev Thank you for listening to me! Wishing you high performing applications