The more I read that post on the "Road To Failure" blog, the less sleep I want to get.
You simply can't scale when you have to deal with network latency. More cores, RAM, and disks talking to each other will always be much, much more efficient than waiting for little packets to cross the ethernet.
How many Googles or Yahoos are there? Like, 5. Let them do whatever broken things they want -- it works for them... for now. It's still expensive, probably just as much as "big iron". Not to mention the countless engineer hours and hosting/electricity costs for their "scale out" systems. It's what happens when you let a bunch of ivory tower PhDs solve real engineering problems.
In the end, the rest of us serious enterprise engineers will allow Oracle, Microsoft, and the people who have been doing this for 30 years to optimize their code to run on multicore mainframes
That article you linked to is one of the craziest things I've ever read. RDBMSs are powerful enough to do anything -- but so many engineers are too lazy to learn proper SQL.
If you need to handle more transactions or queries, buy a bigger box. You can have hundreds of TB in a nice enterprise-grade server. If you HAVE to run your data across multiple machines, just spend some time and actually *think* about what data you want where, and then write a little code to send it to the right database.
Even Google used a ton of MySQL boxes for a long time to deliver their searches. Scale-out architecture is a lot more of a lie than its proponents make it out to be, check out this post on CodingHorror.
Eventually, all these new "databases" will need to actually be used in a *real* environment, and they'll have transactions, SQL support, indexing, a good UI, and everything else. All MS or Oracle need to do is a little tweaking to make their multi-box configurations more robust, and they'll crush everything else.
No man is an island if he's on at least one mailing list.
