Terrastore - Scalable, elastic, consistent document store | Hacker News Reader

Terrastore - Scalable, elastic, consistent document store

code.google.com

37 pointstyler14 comments

The description given on their Google Code page seems a bit sensationalistic to me. The author has a bit more reasonable look at the features and tradeoffs here: http://sbtourist.blogspot.com/2009/12/terrastore-and-cap-the...

Call it "marketing" ;) Joking aside, the article you mentioned is linked on the home page for that exact reason: put things in the correct context.

No insult intended at all. It was just that as I was reading down that list my brain was screaming "CAP! CAP!". I was glad to find that link there. Looks like an awesome project.

Sure, I was just kidding. Thanks for your kind words, hope you'll find it useful.

It looks like the complement to CouchDB: it actually implements all the hard distributed stuff that CouchDB never delivered on, but doesn't implement any of the fancy stored procedure stuff that CouchDB actually pulled off.

If only I could have the chocolate and the peanut butter together...

Don't worry, 2010 is gonna be the year CouchDB lands the cluster-awareness (among other things). We've been moving forward steadily since 2005 and we're not stopping anytime soon.

Terrastore already implements lightweight "stored procedures" in terms of server side updates: functions which atomically update a value based on some custom logic.

Future versions will provide map/reduce functions as well, probably in both Java and Javascript.

I wonder how much of that he got almost-for-free just by using Terracotta?

Terracotta itself is in effect a transparent ACID object database where collections can be larger than RAM and objects can be accessed efficiently on one machine or shared and updated in implicit transactions.

So like, all of the hard stuff. Implementing a protocol wrapper around Terracotta should be cake. Of course, it does force the end user to learn how to administer Terracotta, which non-trivial in it's own right.

Clustering and storage are completely based on Terracotta, and this is great: relying on a rock solid technology gives you the freedom to focus on more advanced features, such as communication, partitioning, data manipulation and querying.

Which is the same reason Voldemort and Riak rely on third-party storage, or HBase relies on Hadoop.

In fact Terastore is based on Teracotta is one of the most important aspects. This makes us comfortable to see similar performance in numbers related to clustering, storage. Same is true for MemchacheDB, Voldemort and Riak as Sergio mentioned.

Exactly: clustering and storage are more and more "commodity" features, as I see it. The most important thing is going to be what you build upon them, the real value you provide to the user: is data partitioned or replicated? What data manipulation features are provided? And so on ...

At first glance it seems like a java version of couchDB on top of a clustering system.

Sounds like this is a lot more closer to MongoDB than CouchDB. Either way, I warmly welcome this to the document store world

Reply on news.ycombinator.com