The Solid State Future.

laserlike.com

10 points·mspeiser··5 comments

Chart combining pricing with sequential and random I/O performance data: http://perspectives.mvdirona.com/2009/04/18/SSDVersusEnterpr...

The non-enterprise SSDs (including Intel's -M line) are now at under 40% of the price (i.e. USD 5/GB), and can probably deliver 80% of the performance for realistic workloads.

I'd also guess that by this time next year we'll be seeing the consequences (and USD 2 or 3/GB prices, if the demand doesn't drive prices up instead).

The biggest change will not be the HDD -> SSD transition, but rather the SSD -> MRAM transition or, equivalently, the point at which there is suddenly no speed difference between RAM and secondary storage.

At that point, there WILL be no secondary storage. Or, rather, CPU cache will become very large and be the primary storage and (M)RAM will be the secondary storage.

In either case, given that CPU cache is (designed to be) transparent, the whole computing landscape will be irrevocably altered.

Think about it... what would a computing environment look like if it was based around a system in which all your data and code was just as instantly addressable and accessible as all your other data and code, and it was just ONE address space?

Summary: solid state disks cost more per GB than spinning disks. But they make computers go faster, so you need fewer computers, so you may save overall.

In fact, it's only in very specialized situations where having faster disks means you need fewer computers.

More than that. A two orders of magnitude kick to the backside of the memory hierarchy necessarily has a ripple effect (e.g. http://news.ycombinator.com/item?id=557961).

It's not only the savings in current applications - it's the new ones that are made possible, or cost-effective. Note the tail end of the piece, and comment #6 - avc.com's ears are pricked up ;-)

Reply on news.ycombinator.com