Scalable Web Applications with a Transactional Key Value Store
summary
Scalaris is a distributed Erlang key value store that provides transactional, strongly consistent updates so a wiki can atomically update revisions, backlinks, categories and counters. The original naive mapping stores giant single keys for page lists and backlinks which makes edits and creates slow (edits ~200 ms, creates 2–2.5 s). Hash partitioning those lists into many shards yields edits ~50 ms and creates <100 ms, and throughput scales roughly linearly as nodes grow from 4 to 24.
AI-written summary · about
chapters (8)
- 0:00Introduction to Scalable KV Store
- 2:58Key Value Store Features
- 4:47Wiki Application Design
- 8:28Data Model Challenges
- 11:15Partitioning for Scalability
- 12:42Application Architecture
- 15:10Scalability Benchmarks
- 15:55Q&A Session
related talks
24mUsing a Distributed Key-Value Store
42mScaling ACID SQL
49mSpanner - fully managed horizontally scalable relational DB with ACID transactions
19mBuilding KVStore for ML Workloads
23mSRFI-167, SRFI-168 and the functional store
29mWhen Web Scale Doesn't Make Sense
59mDistributed Abstractions
41mTransactions: myths, surprises and opportunities