Being an old-school OSS’er, MySql has been my go-to DB for data storage since the turn of the century. It’s great, I love it (mostly) but it does have it’s drawbacks. Largest of which is it’s now owned by Oracle which does a HORRIBLE JOB of supporting it. I have personal experience with this, as the results of a recent issue with InnoDB and MySQL.

In the mean time, some of the hot-shot up-and-commers in another department have been facing their own data processing challenges (with MySql and other DB’s), and have started to look at some highly scalable alternatives. One of the front-runners right now is Apache’s Cassandra database project.

The synopsis from the page is (as would be most marketing verbiage) very encouraging!

The Apache Cassandra database is the right choice when you need scalability and high availability without compromising performance. Linear scalability and proven fault-tolerance on commodity hardware or cloud infrastructure make it the perfect platform for mission-critical data. Cassandra’s support for replicating across multiple datacenters is best-in-class, providing lower latency for your users and the peace of mind of knowing that you can survive regional outages.

This sounds too good to be true. Finally a solution that we might be able to implement and grow, and one that doe not have the incredibly frustrating drawback of InnoDB and MySql’s fragile replication architecture. I’ve found out exactly how fragile it is, despite have a cluster of high-speed specially designed DB servers, the amount of down time we had was NOT ACCEPTABLE!).

With a charter to handle ever growing amounts of data and the need for ultimate availability and reliability, an alternative to MySQL is almost certainly required.

Of the items discussed on the main page, this one really hits home and stands out to me:

Fault Tolerant

Data is automatically replicated to multiple nodes for fault-tolerance. Replication across multiple data centers is supported. Failed nodes can be replaced with no downtime.

I recently watched a video from the 2011 Cassandra Conference in San Francisco. A lot of good information shared. This video is available on the Cassandra home page. I recommend muscling through the marketing BS as the beginning and take in what they cover.

Job graph for ‘Big Data’ is skyrocketing.

Demand for Cassandra experts is also skyrocketing.

Big data players are using Cassandra.

It’s a known issue that RDBM’s (ex. MySql) have serious limitations (no kidding).

RDBM’s generally have an 8GB cache limit (this is interesting, and would explain some issues we’ve had with scalability in our DB severs, which have 64GB of memory).

The notion that Cassandra does not have good read speed, is a fallacy. Version 0.8 read speed is at parity of the already considered fast 0.6 write speed. Fast!?

No global or even low-level write locks. The column swap architecture alleviates the need for these locks, this allows high-speed writes.

Quorum reads and writes are consistent across the distribution.

New feature of local LOCAL_QUORUM allows quorums to be established from only the local nodes, alleviating latency waiting for a quorum including remote nodes in other geographic locations.

Cassandra uses XML files for schema modifications. In version 0.7 provides new features to allow on-line schema updates.

CLI for Cassandra is now very powerful.

Has a SQL language capability (yes!).

Latest version provides much easier to implement secondary indexing (indexes other than the primary).

Version 0.8 supports bulk loading. This is very interesting for my current project

There is wide support for Cassandra in both interpreted and compiled OSS languages, including the ones I most frequently use.

CQL Cassandra Query Language.

Replication architecture is vastly superior to MySQLs transaction and log replay strategy. Cassandra uses an rsync style replication where hash comparisons are exchanged to find which parts of the data tree a given replication node (that is responsible for that tree of data) might need updating, then then transferring just that data. Not only does this reduce bandwidth, but this implies asynchronous replication! Finally! Now this makes sense to me!!

Hadoop support exists for Cassandra, BUT, it’s not a great fit for Cassandra. Look into Brisk if Hadoop implementation is desired or required.

Division of Real-Time and Analytics nodes.

Nodes can be configured to communicate with each other in an encrypted fashion, but in general inter-node communication across public-private networks should be established using VPN tunnels.

This needs further research, but it’s very, VERY promising!

NEXT: “Cassandra and Big Data – building a single-node ‘cluster’“