Floppy-to-Disc Like Disc-to-Solid State
One of our consultants, Dan Souk got me thinking about the impact Solid State Discs (SSD) might have on our technology world. And the possibilities are indeed exciting – I can’t wait for these to become economical in quantities that will support widespread adoption over traditional physical mechanical platter / read-head disks. SSDs have multiple advantages starting with the obvious one of no moving parts. But the most important advantage is raw speed which will likely lead to a change in how enterprise applications are architected and deployed.
Why is this a big deal? Consider the ability to truly utilize a modern CPU’s capabilities. If you’ve ever looked at your PC’s system monitor, you’ve undoubtedly noticed the CPU loafing along at some very low percentage utilization (10% or less) while your I/O operations are pegged at near 100%. Solid-state discs will allow CPUs to earn their keep by working much harder, and making your everything go must faster. Performance improvements are dramatic – I/O wait times are cut by a factor of 250x allowing for an overall increase of 80x more I/O transactions per second.
One of the biggest constraints with business applications is the latency of disk access. Have you ever pondered the impact on database design if disk latency effectively goes to zero (it will probably not be zero, but for sake of argument it will be pretty close when compared to moving physical heads and platters)? Consider:
- Do we really need things like stored procedures, from a physical point of view?
- Do query execution plans really matter when the entire database is milliseconds away, practically in memory?
- Do you have to partition tables to optimize performance?
The answer to these questions is probably still yes, but not as much as before. It’s exciting to see this develop, because it should lead to simpler, less complex systems.
And as systems go faster (much, much faster) fewer servers will be needed to run large enterprise class applications. The effect is dramatically lowered cost per transactions due to vast increases in speed. Hardware costs will drop and system reliability will go up. Most are aware that discs, even in a RAID configuration, are a weak link in system reliability. So how does a SSD MTBF (Mean Time Between Failure) of 228 years sound? When deployed as a redundant pair the possibility of data loss virtually disappears due to a MTBF of 310,000 years (I never worry about these kinds of time frames).
How will SSDs change your world?
I welcome your comments.