I am yet to see any evidence that RethinkDB's approach is anything but a niche for those who need to squeeze an extra 5-10% that advances in filesystems and flash firmware cannot.
The interesting part comes from being able to execute existing operations in fundamentally different ways. When you're not restricted to sequential reads you can implement a huge chunk of database internals differently - different concurrency models, different implementations of certain joins, etc. This is where the bulk of the performance improvement comes from.
Once you do this, you realize that a traditional bottleneck (I/O) is completely blown away. Suddenly, all kinds of things nobody ever bothered to optimize become performance bottlenecks (for example, the query plan is traditionally interpreted, not compiled to native code - not good enough once you get an I/O subsystem that's 250 times faster than what you used to have).
Once you get rid of the software bottlenecks, you realize that you're limited by the bus. Since SSDs are highly parallelized devices, you have to change the way you interleave operations for optimal performance. That alone is another order of magnitude.
In three years or so, the flash manufacturers will put chips on the SSDs that will allow you to send down programs to them (not unlike the shaders in video cards). That will likely be a bigger revolution than SSDs themselves. Where we are now is just the beginning.
You're right that we haven't shown any evidence that our approach is anything but a niche. We will. This stuff just takes a long time to build.
Actually, System R (IBM's pioneering relation database from the mid 1970s) compiled query plans down to machine code.
This sounds crazy. SSDs have all this firmware and processing power to make them behave like hard drives. If we want to drive the underlying storage directly we might as well rip all that out and let the OS drive the flash directly from the main CPU. Why would we want to be sending programs into the black box? What can the firmware do that the main CPU couldn't if we stopped attaching flash as SATA and instead just created a direct connection that can DMA back and forth to main memory?
Some Linux developers have made that point, that they would rather be implementing their own filesystems directly on the flash than dealing with a firmware that is trying to make it appear that the SSD is a hard drive but that then needs TRIM to work around the fact that that is not true. These firmwares are nice for Intel by letting them turn a commodity (flash) into a differentiated (and higher margin) product based on the code in the firmware. Considering Linux already has filesystems that can operate directly over flash and that better ones can be written I'd say we would be better off just getting direct access to the flash.
Flash is 10x slower than RAM (in an ideal case), and even then, we haven't had the need to introduce 'shader like' features into RAM - yes, because historically we haven't stored all that much in it, but even with modern 64Gb configurations, same thing...
Having said that, it's also possible that the actual 'software overhead' is pretty much negligible when compared to disk access times.
Would love to learn more on this.
Upgrading our current hardware with only $10k worth of SSDs will add at least a year or two to the life of our SQL Servers.