for Information Management Blogs
APR 23, 2009 1:23pm ET

Blogroll

Make the Whirl Go Away

Print
Reprints
Email

One of the more interesting “creeping” technologies of 2009 has to be the introduction of solid-state hard disks at the low and high ends of the market. On the low end, more and more laptops don’t spin up with that familiar sound and, once you get past your initial panic, you find they boot faster, use less energy and stay cooler than your old machine. You’re already used to this behavior with your iPod. Tiny 2.5” and 1.8” drives are commonly available up to hundreds of gigabytes, and with retail prices dropping toward $2 per GB, they’re becoming common and practical rugged little green devices.

The high end of the market is something else. I got a pitch this week from Texas Memory Systems touting their new RamSan 620, which claims to be the world’s largest, fastest solid-state drive (SSD) flash system. It’s five terabytes, 250,000 sustained I/Os per second (for random reads and writes) with latency claimed to be 60 times less than hard drive disks (HDD). It takes up two rack spaces, draws 375 watts of electricity and costs $220,000.

Texas Memory claims the system replaces as many as 500 HDD drawing 7,000 watts and, with supporting gear included, a cost upwards of a half-million dollars.

I’m no hardware geek so I called up Woody Hutsell, who is Texas Memory’s president, a patient man who walked me through many of the details. For starters, Hutsell has worked with a lot of RAM devices, which are faster than drives of any kind, though he sees flash SSD as a bridge between RAM and slower bulkier HDD arrays. The opportunity he sees is in improving storage devices so as to let application performance catch up with much greater strides in computer processing speed.

“We’ve built things that never made it to market, this is probably our third flash controller so you could say we’re just getting better at it.” Plus, over his nine years in the business, he’s seen high performance SSD memory fall from about $9,000 per GB to around $40. You don’t have to be a hedge fund analyst to see where this is leading.

But $40/GB is a lot more than HDD costs today, so what about those cost numbers? “The acquisition cost is probably the biggest non-FUD factor,” says Hutsell. “You could look at it and say, if I needed five terabytes of fiber-channel disk, I could buy 40 146-gigabyte disks loaded for $1,000 per drive. If you just compare $40,000 to $220,000 it’s not a hard decision. But then the architect says, 40 drives at 300 I/Os per drive are going to give me 12,000 I/Os per second in the best case, assuming I have a great controller. If I need more I/Os I need to start striping a five-terabyte data set across 3X as many drives depending on my goal. Now I really am more expensive on a cost of solution basis.”

Boiling all that down, the SSD community would rather talk about cost per I/O than cost per gigabyte. “If everybody thought in cost per I/O my job would be easy but the reality is they don’t. What I’m suggesting is that in cost per solution we can compete effectively.”

Woody Hutsell does not see any particular tipping point or imminent end for HDD or even tape drives for that matter. It’s more of a nibbling away that is gaining traction in telecom, financial services, government, e-commerce and technology for uses like online transaction processing and data warehousing.

“There are scenarios where a customer, let’s say, has an Oracle 10 database and wants better data warehouse performance. They might say, one way to get better performance is to add a solid-state disk and another way is to go to Exadata or to a Netezza,” Hutsell told me. “It could be that we’re the method they use to accelerate what they have and then they can use that as a comparison point against parallel processing benefits of those architectures. There’s no reason Oracle or Netezza couldn’t benefit from flash SSD to accelerate their tools either, though we’re not working with them today. It’s not purely competitive but it is a way people could accelerate their incumbent architecture with very little distraction.”

I am curious what all you database managers and administrators think of all this. There are pluses and minuses to SSD we didn’t touch on here. Even HP tap dances around the issue of SSD in their marketing. So sound off and set me right.

Filed under:

Advertisement

Comments (1)
We are using an earlier TMS SSD device on our DW infrastructure. For us, to get better performance we could have upgraded the memory on all of our Oracle RAC nodes, (or bought bigger server nodes), at some cost. Instead with a shared SSD device, we can use the fast storage for Oracle TEMP workspace every time processing spills from server memory. Some of our load jobs and large queries speed up from 40 to 70%.
Posted by Roland A | Wednesday, April 29 2009 at 11:31AM ET
Add Your Comments:
You must be registered to post a comment.
Not Registered?
You must be registered to post a comment. Click here to register.
Already registered? Log in here
Please note you must now log in with your email address and password.

Blog Archive for Jim Ericson

Lighting the Borders of Information Management
Data Governance in Context
Big Data: Bring Popcorn
In IT, More Needs to Be Less Too
Mainstreaming BI

More from Jim Ericson »

Blog Index »

Twitter
Facebook
LinkedIn
Login  |  My Account  |  White Papers  |  Web Seminars  |  Events |  Newsletters |  eBooks
FOLLOW US
Please note you must now log in with your email address and password.