1. A single character change led to a 3.3x performance boost at Segment's internal tracing storage infrastructure, TraceDB.
This is the story of how a tiny tweak made a massive impact. Keep reading to learn more! 🚀
2. TraceDB's sorting algorithm is designed to sort tracing spans by trace ID and flush them to disk in an LSM tree during compaction. But the team at Segment noticed it was too sluggish when handling incoming queries.
3. They discovered that it was only doing work on one CPU core while others were idle. Time for some root cause analysis!
Stanford announced the Alpaca 7B model. It's a fine-tuned model based on Meta's LLaMA. It performs similar to ChatGPT but is lightweight enough to run on consumer devices. GPT-like models are getting cheaper and more accessible really fast.
You loved my tweet on Stack Overflow's monolith. Now, let’s explore the opposite end of the spectrum: Netflix API architecture evolution.
Discover how Netflix API evolved through 4 stages.
1. Monolith
Like everyone else, Netflix started with a monolithic architecture where the entire application was packaged and deployed as a single entity.
2. Direct Access
To improve team autonomy and speed, the monolith was split into microservices. Each client made direct requests to the microservices. But with hundreds of them, exposing everything wasn't ideal.
1. Let's talk about a memorable challenge in running a massive online game years ago - the battle against DDoS attacks in online poker!
2. Why would anyone DDoS a poker game you might ask?
Here's one reason: to block anyone from sending continuation bets resulting in an automatic forfeit. The ill-gotten gains were then offloaded to a thriving black market.
3. So, how does a DDoS attack work?
The goal is to disrupt the normal flow of traffic to a service by sending a large volume of attack traffic from compromised devices called botnets.