Hacker News new | past | comments | ask | show | jobs | submit login
Accumulo: NSA's Apache-licensed BigTable-based key-value store (wikipedia.org)
79 points by gojomo on June 6, 2013 | hide | past | favorite | 5 comments

> In June 2012 Congress's Senate Armed Services Committee (SASC) released the Draft 2012 Department of Defense (DoD) Authorization Bill, which included references to Apache Accumulo. In the draft bill SASC required DoD to evaluate whether Apache Accumulo could achieve commercial viability before implementing it throughout DoD.[5] Specific criteria were not included in the draft language, but the establishment of commercial entities supporting Apache Accumulo could be considered a success factor

This was actually pleasantly surprising to read...the government actually wants to see if its software is useful and competently designed, rather than just buying it because the contractor has influence and no one will know how crappy it is anyway since it'll be classified.

When Accumulo was developed in-house at the NSA it was called Cloudbase. When it became part of the Apache project it was renamed Accumulo.

I've worked with both HBase and Cloudbase/Accumulo. HBase tooling is more mature than Accumulo's but performance characteristics are about the same. The extension model for Accumulo - for custom iterators and aggregators - is pretty easy to get the hang of right away. (In fairness I never looked into extensibility for HBase.)

Both systems play well with Hadoop. Accumulo is a little bit more difficult to use with Oozie.

Accumulo's built in monitoring is decent.

Well.. uhh... Now we know what this project was specifically used for.

But seriously, usage concerns for the project notwithstanding, it looks like a cool K/V storage system. I wonder how it performs compared to similar systems and how it handles failure.

http://www.pdl.cmu.edu/SDI/2013/052013-a.html is a rather interesting, recently published talk on the subject of scalability of Accumulo

It operates similarly to HBase but has built in security levels and easy extensibility for custom iterators and aggregators.

Failure I cannot speak to since I really only develop against a local cluster that I blow away often.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact