Skip to main content

AtlasDB: ACID Transactions for Your Favorite Key-value Store

Ari Gesher (Kairos Aerospace), Danielle Kramer (Palantir Technologies)
Hadoop & Beyond Gramercy Suite
Average rating: ***..
(3.67, 3 ratings)
Slides:   1-PDF 

AtlasDB is a bolt-on layer for a key-value stores (distributed or otherwise) that implements MVCC and guarantees ACID properties for eventually-consistent data stores. Palantir is currently vetting this software for possible open source release for use as a community tool, bringing transactions to a key-value store near you.

In this talk, we’ll take a look at the protocol used to implement the transactions, talk about the performance tradeoffs from using transactions, and look at the transactions API it offers.

The Palantir team that built AtlasDB refined Google’s transactional protocol called [Percolator](http://research.google.com/pubs/pub36726.html). In the process, they built lighter weight transactions with less performance overhead than the original Percolator implementation. In the talk, we’ll walk through the details of the protocol and show a proof of how it makes its guarantees.

AtlasDB implements multi-version concurrency control (MVCC) that allows consistent and writes reads from an [eventually consistent](http://en.wikipedia.org/wiki/Eventual_consistency) distributed key store, like Cassandra, MongoDB, HBase, etc., removing an ever-present challenge to using these super-scalable data stores to back for more complex operations.

AtlasDB is agnostic to the key-value store used; we’ll look at the plugin architecture needed to implement an adaptor to connect the transactional layer to the key-value store. AtlasDB then exposes an easy-to-use API that includes transactional primitives and error-checking.

Finally, we’ll look at performance benchmarks to quantify the overhead imposed by using transactions and the relative performance of different key-value stores when using AtlasDB.

Photo of Ari Gesher

Ari Gesher

Kairos Aerospace

A software/systems engineer with a lot of experience building big, real-world systems.

Danielle Kramer

Palantir Technologies

Danielle Kramer leads the Backend Infrastructure Engineering Team for the Palantir Gotham Data Fusion Platform. An accomplished software engineer in her own right, Danielle herds the cats that build the low-level data plumbing that makes the magic Palantir’s platforms possible.

Sponsors

Sponsorship Opportunities

For exhibition and sponsorship opportunities, contact Susan Stewart at sstewart@oreilly.com

Media Partner Opportunities

For information on trade opportunities with O'Reilly conferences email mediapartners
@oreilly.com

Press & Media

For media-related inquiries, contact Maureen Jennings at maureen@oreilly.com

Contact Us

View a complete list of Strata + Hadoop World 2013 contacts