Concurrency + Distribution = Scalability + Availability (...)
summary
Erlang/OTP is an enabler not magic: prototypes and numbers decide. When nodes are fully connected performance and monitor overhead start limiting clusters around 50–70 nodes, so production scales use the Dynamo principle with 64 vnodes and tools like riak_core for consistent hashing gossip hinted handoffs, enabling dynamic add/remove and redundancy. Examples: an event store built on riak_core hits millions of reads/sec; midsize clusters handle ~20–40k TPS; Phoenix shows >2M TCP connections on one machine. Tradeoffs are explicit: consistency vs availability, retry modes, and backpressure choices.
AI-written summary · about
chapters (17)
- 0:00Start
- 5:20Node Types
- 6:11Logic Nodes
- 8:11Pros
- 9:22The Dynamo Principle
- 12:22Consistent Hashing
- 14:38Peer-to-Peer Approach
- 15:12Network Protocols
- 17:30The Node Interfaces
- 19:21Fault Tolerance
- 20:29Resilience
- 20:59Reliability
- 25:51Primary Primary Replication
- 26:31Network Partitioning
- 29:52Exactly Ones Request
- 35:46Stress Testing
- 39:41Monitoring and Pre-Emptive Support
related talks
01h 00mFrom Concept to Reality: Solving Enterprise Challenges
49mThinking in a Highly Concurrent, Mostly-functional Language
51mScalability Q&A
01h 00mHigh Optionality Programming: Architectural Choices That Mitigate Technical Debt
40mDistributed, Eventually Consistent Computations
49mJourney with Aerospike: Architecture Best Practices
46mDigital scale-up architectural patterns
26mGrowing Pain: Scaling to 10K Workflows per Week