I work for a company that operates in the energy industry. We utilize both RDBMS and "NoSQL", both have their purposes that they fit in well. We store customer account and configuration data in Postgres, and use Cassandra to store time-series statistics and high write volume data.
I have a background in data warehousing in both Oracle and SQL Server, and was part of the decision to use a polyglot persistence model. I've got at least a decade's worth of experience in the DW world, and more as a general developer before that, so I like to think I've got a relatively credible background in a variety of data stores.
I haven't looked at Mongo much - it's durability concerns and the write lock stuff pushed me away from it early on (I don't mean to disparage it, but that was where it was at when I evaluated it), but Cassandra's configurable consistency levels and operational story at a cluster level are what sold us for our time-series data (that, and the ability to construct a sparse timeline and multiplex reads/writes). For anything we need flexible querying with, we push it into specialized Postgres dbs.
The level of willful ignorance and vitrol in this thread is kind of amazing. Most of the really experienced DW guys I know are all looking at HBase, Cassandra and others because they fit a niche that we've all been looking for in certain data sets at really large scale. It doesn't mean we're ditching our relational data stores, it just means we're augmenting them with other tools because they fit the job at hand. To suggest that one tool is absolutely perfect for every scenarios seems a little short-sighted to me, possibly driven out of inexperience. I don't mean that as an insult - I know a lot of guys who've been working on the same data sets for 30 years who really do just need the one tool - however, you've got to realize there are other data sets and problems for which your hammer just won't fit.
I have a background in data warehousing in both Oracle and SQL Server, and was part of the decision to use a polyglot persistence model. I've got at least a decade's worth of experience in the DW world, and more as a general developer before that, so I like to think I've got a relatively credible background in a variety of data stores.
I haven't looked at Mongo much - it's durability concerns and the write lock stuff pushed me away from it early on (I don't mean to disparage it, but that was where it was at when I evaluated it), but Cassandra's configurable consistency levels and operational story at a cluster level are what sold us for our time-series data (that, and the ability to construct a sparse timeline and multiplex reads/writes). For anything we need flexible querying with, we push it into specialized Postgres dbs.
The level of willful ignorance and vitrol in this thread is kind of amazing. Most of the really experienced DW guys I know are all looking at HBase, Cassandra and others because they fit a niche that we've all been looking for in certain data sets at really large scale. It doesn't mean we're ditching our relational data stores, it just means we're augmenting them with other tools because they fit the job at hand. To suggest that one tool is absolutely perfect for every scenarios seems a little short-sighted to me, possibly driven out of inexperience. I don't mean that as an insult - I know a lot of guys who've been working on the same data sets for 30 years who really do just need the one tool - however, you've got to realize there are other data sets and problems for which your hammer just won't fit.