"Let’s say you have 10 monster DB servers and 1 DBA; you’re looking at about $500,000 in database costs."
I wonder what he thinks is a "monster db server", and considering he included the DBA in the price, is this the price per year or what?
Having recently set up a dual E5620 with 48GB of RAM and 8 SSD drives(160GB each) with a 3ware controller as well for just shy of 10K USD, I guess my understanding of "monster" is quite different. For 13K USD the same server would have 96GB of RAM.
The numbers in the article are strangely inflated. In addition, as if the need for DBA disappears when you've simply changed your data storage software. Somebody still has to know how it works and manage it.
If you don't need a 50 node cluster because your RDBMS is pulling down big numbers, then you don't multiply the cost of the RDBMS solution by 50 either.
The numbers posted here are pretty reasonable. 37Signals spending $7,500 on disks isn't outrageous. That's less than the cost of a single developer integrating a different solution over a few months. How long has Digg been working on this transition and how many employees did it require? They've probably spent a fortune. Just not on hardware.
It is arguable that the need for a DBA does disappear with the change in the systems. The base approach of the relational model is that there is a professional DBA who makes sure that the mission-critical data is available in a format that multiple applications and multiple department can use directly (the type of data is know but the ordering isn't set). Thus DBA is "hat" that most programmers can't "wear", especially since a large portion of programmers don't understand the relational model.
On the other hand, Nosql and object-databases allow a programmer to just stuff data into the a data-store without worrying about a cohesive datamodel. If we consider this as mission-critical data that multiple departments of a large organization would want to see in multiple forms, then we can find many ways that the approach of "just save this array of values" produces serious problems.
But there are many applications where these problems don't appear. Diggs seems like it could get away with doing nosql. A health-record site seems like it could not do nosql since it ultimately is going to want ACID-and-beyond in its data model.
Good comment, but I wouldn't quite say that using a non-relational database frees you from having to think about a cohesive data model. It might be more accurate to say that for some data models non-relational stores are a more natural fit which frees you having to think about how to force your model into the wrong container.
It's hard to find the right adjective for what's distinct about the relational model.
The relational model is a fantastic model of data independent of application. It can even be a great model for an application using the same data in different ways.
But this approach clearly has a cost. In ways, there's the question - is this an application with a company built around it or a company with a application built around it? Digg and Google are applications with companies built around them. Here the RDMS model doesn't make sense.
Could be, though he speaks about MySQL, PostgreSQL as well, and in other places about MS SQL, mostly hammering the license cost, which is obviously huge.
One more thing I'd add is I have no clue who these upset DBAs are and who is thinking Stump & Co. are dumb. Everyone making these sql/no-sql blog posts seems like they're starting a war with made-up enemies.
I wonder what he thinks is a "monster db server", and considering he included the DBA in the price, is this the price per year or what?
Having recently set up a dual E5620 with 48GB of RAM and 8 SSD drives(160GB each) with a 3ware controller as well for just shy of 10K USD, I guess my understanding of "monster" is quite different. For 13K USD the same server would have 96GB of RAM.