RAMP-TAO: Layering Atomic Transactions on Meta's Online TAO Data Store
summary
Describes RAMP-TAO, a layered protocol that adds atomic visibility to Meta’s TAO by using the existing refill library for limited multiversion metadata and marking recently transactional keys; it finds atomic-visibility violations occur about 1 in 1,500 naive batched reads and can last up to 30 seconds, motivates needing fixes, and reports feasibility: bounded memory overhead, only ~3% of data is transactional, and over 99.9% one-round reads with tail latency similar to baseline.
AI-written summary · about
chapters (8)
- 0:00RAMP-TAO introduction
- 2:33TAO data store evolution
- 5:21Atomic visibility violations
- 9:34RAMP on TAO challenges
- 13:02RAMP-TAO design
- 17:39Performance impact
- 18:21Lessons from guarantees
- 20:06Conclusion
related talks
17mMako: Speculative Distributed Transactions with Geo-Replication
47mBeing transactional in 2019 - what does this actually mean?
14mRetrofitting High Availability Mechanism to Tame Hybrid Transaction/Analytical Processing
01h 13mHTAP made live — from the engineering perspective
43mBeing Transactional in 2019 — What Does That Actually Mean?
37mGeo-Replicated Transactions in 1.5RTT
23mCache Made Consistent – Cache Invalidation Might No Longer Be a Hard Thing in Computer Science
49mThe Paradigm Shift - Understanding the Move To Distributed Databases