Distributed Abstractions

on YouTube ↗59m

" summary

Describes Mastry’s tree and the Silo database built on it: in-memory key/value storage using a compare-and-swap header with lock bit and version per node, readers that do no-writes-for-reads (optimistic reads that validate the header), writers that lock and bump versions, and a three‑phase commit that locks, verifies read sets, then installs writes. Implements one worker thread per core and assumes one-shot requests. Shows high scalability across cores, outperforms partitioned schemes when cross-partition transactions rise. Also generalizes into Stow transactional objects and discusses epoch GC and encapsulation.

AI-written summary · about

" chapters (12)

1. 0:00Start
2. 1:24Multicore Issues . Good performance when synchronization is required . Scalability
3. 2:23Multicore machines Sort of distributed
4. 6:20Concurrency Control
5. 13:27Transactions begin % do stuff: run queries % using insert, lookup, update, delete, % and range
6. 14:04Silo Database is in primary memory . Runs one-shot requests
7. 15:13Running Transactions
8. 23:10A vision for concurrent code Apps run transactions
9. 27:12Transaction-aware datatypes Provide ops for user code .E.g., lookup, update, insert, delete, range . Record reads and writes via platform
10. 29:04Example: deleting a key - Transaction that deletes k commits . Removing a key is expensive
11. 33:33Performance Silo: 7000 lines of code - STO-Silo: 3000 lines of code . Uses hash tables and trees
12. 34:58Observations . This work depended on abstract data types . Better performance by taking advantage of

" related talks

• 29mModern Database Architecture: From SQL to NoSQL & Distributed SQLAndrei Manakov · Site Reliability Engineering (SRE) 2025
• 52mThe Computer Science behind a modern distributed data storeMax Neunhöffer · Devoxx Belgium 2017
• 41mTransactions: myths, surprises and opportunitiesMartin Kleppmann · Strange Loop 2015
• 52mTransactions: Myths, Surprises and OpportunitiesMartin Kleppmann · Code Mesh 2015
• 19mBuilding a Performant Configuration-Store LayerGuarav Agarwal · FoundationDB Summit 2019
• 16mScaleDB: A Scalable, Asynchronous In-Memory DatabaseSimon Peter, Lorenzo Alvisi, Syed Akbar Mehdi, Deukyeon Hwang · OSDI 2023
• 46mLost in transaction? Strategies to deal with (in-)consistency in distributed systemsBernd Rücker · Full Stack Fest 2018
• 34mWhat We Talk About When We Talk About Disk IOAlex Petrov · ScaleConf Colombia 2018