YC
Things That Don't Scale, The Software Edition – Dalton Caldwell and Michael Seibel
@ Y Combinator
02/17/22
Takeaways
MS
Michael Seibel
02/17/22
@ Y Combinator
If you want to test your product before launching, think of it like a house with broken pipes; instead of checking every pipe, just turn on the water to see what needs fixing.
MS
Michael Seibel
02/17/22
@ Y Combinator
Paul Buchheit emphasizes the '90/10 solution', which means finding a way to achieve 90% of the benefit with only 10% of the work.
MS
Michael Seibel
02/17/22
@ Y Combinator
Paul Buchheit, the inventor of Gmail, built it as a side project at Google, focusing on doing things that don't scale, which ultimately led to a product used by 1.5 billion people.
MS
Michael Seibel
02/17/22
@ Y Combinator
When Gmail was first developed, Paul Buchheit created a makeshift email interface using Google Groups, which allowed him to read his emails and led to the development of essential email features.
MS
Michael Seibel
02/17/22
@ Y Combinator
The initial Gmail invite system was not a clever growth hack; it was a necessity due to limited server space, forcing them to limit access until they could scale.
MS
Michael Seibel
02/17/22
@ Y Combinator
In the early days of Facebook, they scaled by creating separate database instances for each school, allowing them to manage growth without overwhelming a single database.
MS
Michael Seibel
02/17/22
@ Y Combinator
Twitch faced significant traffic spikes, sometimes 20 times higher than normal, which required them to implement hacks like turning dynamic pages into static ones to handle the load during popular streams.
MS
Michael Seibel
02/17/22
@ Y Combinator
Friendster calculated the degrees of separation for users in real-time, which caused performance issues, while MySpace simplified this by stating whether someone was in your friends list or extended network, demonstrating a 'do things that don't scale' approach.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
When a major celebrity announces a live stream, a hundred thousand fans can request it simultaneously, causing the video server to crash. We had to find solutions to manage this sudden influx of traffic effectively.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
At Twitch, we learned that the video system doesn't know how many viewers are waiting before a stream starts, but the website does. By communicating this information, we could pre-populate the stream to the necessary video servers before going live.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
YouTube innovated by allowing videos to play in the browser using Flash, eliminating the need for external applications like RealPlayer, which often crashed browsers. This was a significant product innovation at the time.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
At imeem, we used a hack to play music files as video files, transcoding uploads into video formats that had no visual content, effectively creating a video site that only played audio.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
When we faced high costs for streaming live video, we implemented free peering relationships with ISPs to save money. If an ISP refused, we displayed a message to users explaining the situation and encouraging them to contact their ISP.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
To translate our site into multiple languages without incurring high costs, we created a community-driven platform where users could volunteer to translate content, similar to a model used by Reddit.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
Google's original search algorithm struggled to scale as the web grew, leading to a period where users received stale search results. This crisis prompted the development of MapReduce, a system for parallelizing web crawling and indexing.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
The lesson from startups is to focus on creating something people want first, even if it involves non-scalable methods. Once you have demand, you can then work on scaling solutions.
YC
Y Combinator Cast
02/17/22
@ Y Combinator
Don't let the pursuit of perfection hinder progress. It's better to provide something good and iterate on it than to wait for the perfect solution.