Wow – 1TB consumer-grade SSDs hit the street.

Anyone doubting that 2010 will be an inflection point on solid-state storage?   We now have consumer-grade 40GB drive at $115 (talked about here), and a 1TB drive at $3.5K.

http://www.theinquirer.net/inquirer/news/1562474/one-terabyte-ssd-hits-shops

While these are consumer drives, and are missing the things that are needed for enterprise-grade solid state storage that STEC does now (much heaver duty cycle, flat performance over time, larger SRAM to Flash ratios, dual ported interfaces, etc) – it really highlights how fast things are moving, and innovation that occurs in consumer-land can trickle up in the same way that the reverse is true also.

I’m sticking with my bet (admittedly more aggressive than others think) – late 2010 will be the point where solid-state will:

1. match fast magnetic media as measured by $/GB (currently about 10x less efficient)
2. will pass being 300x more efficient measured by $/IOps (currently about 100-200x more efficient)
3. will continue to be 100x more efficient measured by Watts/GB (10,000x more efficient measured by Watts/IOps)

When this happens, will be crazy – only room for SSD (SAS/FC interfaces) and huge slow magnetic media (SATA/SAS interfaces).   Everyone will be looking to see how to maximize this.  

There are second-order effects too – we’re getting close right now to the maximum heat density we can with 15K drives – as we increase the drive density, we can’t use 10/15K RPM SAS/FC, only SATA and SSD (this is why dense – meaning >15 drives per 3U - storage configs generally focus on low-power SATA use cases now).

What people also don’t intrinsically get is that non-volatile storage that sits in between the characteristics of memory, but hundreds (thousands even) times faster than magnetic media has broad transformational characteristics – including things like HPC, BI/DW appliances and all sorts of things.  People have developed entire technology stacks around the idea that you have memory (volatile) and then 1,000,000 times slower you have disk (non-volatile).

Between virtualization (many, MANY more exciting things on that front), broad 10GbE adoption, deduplication showing up everywhere, Nehalem (and it’s successors), and solid-state disk moving mainstream – lots of transformational technology elements – 2010 will be an exciting year!

Comments

[DP]

I think the first ones to buy the more accessible consumer-grade SSDs will be the gamers - could you imagine how fast the OS will work on 2 X SSDs in RAID-0? I'm drooling already, though Apple will be the last to adopt it for their Mac Pro (my current pride and joy) :-(.

[Dpironet]

My thought is that current disks, with moving parts, are the last bastion we have (had!) to take down before going to infiniband. As usual the former technology pays for the new one, thus regular disk as we know them, still have some years to live before completely phasing out. Let see what will happen in 2010 :)

[marc farley]

Chad, 2010 might be the next best bet for an inflection point for SSDs, seeing as how 2009 is almost gone and the leading SSD producer, STEC, has had to deal with disappointment. But I think your bet will still be premature for 2010 where enterprise and cloud storage is concerned because the gap between consumer and enterprise SSDs is considerably wider than the gap between consumer and enterprise disk drives.

Consumer SSDs might match $/GB of fast magnetic, but not of slower SATA drives - that still is a problem for enterprise SSDs.

IOPs will be much, much faster - but not as fast as people project based on SSD vendor claims, which use optimized conditions, as opposed to real world, enterprise or cloud, mixed-workload conditions. It's like measuring the speed of a sports car running on a straight line, as opposed to driving a real road with unexpected curves, hills and some amount of traffic. The expectations for disk drive IOP performance comes from years of experience while the expectations for SSD IOPs comes from data sheets. That will take some time to sort out. Nonetheless, SSDs win the IOPs performance challenge 100 times out of 100.

Of course, that doesn't mean they can be used effectively, which will require software to exercise them as much as possible. It's great to have the performance capabilities of a technology, but its even better to be able to put it to work. This is all vaporware today. Will we see a complete turn around in that situation to make a difference in enterprise storage in 2010? I bet not.

I agree that low latency is an advantage for certain enterprise and cloud applications, but reliability is critical for all those apps. Considering the insane duty cycles expected from SSDs to support sub-LUN virtualization (expect much higher duty cycles than fast disks experience today) - SSDs still have a lot to prove in the field. There will be bumps in the road, like there are for all technologies. That's why I'm betting SSDs will be sold alongside fast rotating disks for several years to come until people can clearly understand what the risks are.

Power consumption looks much better for SSDs as you point out and should continue to be for rotating disks from now until the end of time.

So, my bet is that it will be 2012 before we see enterprise SSDs really take off. That leaves the year 2011 for somebody else to wager on. Any takers out there?