IBM chief engineer talks green storage

On the heels of a billion-dollar investment from IBM in  a new "green" storage program, Clod Barrera, engineer and chief technical strategist, IBM Storage, discusses energy efficiency, power consumption and cooling issues, all of which are becoming hot issues (literally and figuratively) in the storage industry.

What are the 'green' issues particular to storage systems?

Clod Barrera: There is a very large appetite for storage capacity and increasingly a large appetite for quick access to data, so a lot of customers are talking about backing up to disk rather than on tape, and disk is of course much more power hungry than tape. There's more and more talk about online archives. People are talking about petabytes of spinning disk to accomplish this. Disk array controllers and [ storage area network] San switches and host bus adapters for storage will also become more and more power hungry over time as speeds increase.

More on green storage Green storage gains ground Users thrash out green storage Copan users get electricity rebate with MAID arrays Solving storage power and cooling concerns Storage industry ponders power efficiency

One of the habits a lot of people fell into when storage was just getting cheaper by half every year is that they stopped managing storage with real science and real tools. It became easier just to deploy a lot of it so you would be sure not to run out. That was in retrospect a bad habit to fall into. Back in the days of mainframe we had a lot of tools and process to manage storage to as high a capacity point as could be balanced with performance requirements. People got medals for reaching high utilisation with high performance on mainframes 20 years ago, but those practices didn't get adopted in distributed environments. Now, even if I can still afford the [capital expense] of hardware, I can't afford the power -- there's a significant value-add for doing a better management job.

The simple projections of where this goes over time become quite daunting. The trend is clear -- if the storage industry is going to both meet the demands users have for high-performance access to information and not let the power bill get out of control, we're going to have to do some innovation.

Do storage arrays from different vendors really draw different amounts of power? How significant is that?

Barrera: One of the discussions we're having within SNIA is how to get really good numbers for that purpose. The simple answer is yes, of course, since different arrays are designed by different people for different purposes, but we do not have enough of a standard yardstick to determine the goodness or badness of any particular box compared to another. There actually is considerable variability. Currently within SNIA, the first project is to establish a set of yardsticks, what do you measure and how, and also how many do you need. One good one is watts per gigabyte stored, but there are a lot of environments where customers don't deploy on the basis of watts per gigabyte or dollars per gigabyte, they deploy against ops per second. So you probably need a watts per ops per second yardstick for true high-performance transactional environments, and maybe you need a watts per gigabyte per second for streaming environments and maybe a different yardstick for archival environments.

We need to argue this out, and we need to do it quickly because we're better served by experts deciding what the right metrics are instead of using a single number that's not a good basis for comparison. We also ought to have a certification process somewhere so your box can be measured and verified by a third party and then we'll really know.

How realistic is that, though? It's not in a vendors' best interest to have real numbers come out, especially if trying to sell storage hardware.

Barrera: Once everyone knows real numbers, the race will be on to build better products for customers. If you have 20% more efficiency, you can compute what it's worth in dollars for the customer and compute a better margin. I personally think there will be enough social and regulatory pressure so that these yardsticks will come into existence one way or another. My goal is for an organisation that knows how to do this, like SNIA, representing the right entities, to decide what those yardsticks need to be. I think there'll be plenty of help from the rest of the world on the enforcement process.

What trends do you see in storage hardware and data management for green issues over the next year?

Barrera: To me the single biggest rock in the pond early is going to be the notion of measurements and standards. Once measurements are in place, you're going to see significant innovation to drive down power consumption and drive up heat efficiency -- the nice thing about numbers is they encourage a lot of behavior. The technologies that will be important over the next year, however, are all already in place -- things like storage virtualisation, which drives up utilisation, information lifecycle management [ILM], which moves data to lower power media and data deduplication and reduction. It will probably take about a year to get the right measurements in place, another year for best practices and over time products will get better and better. We'll probably have a good place to look back from in five years.

What options does IBM have in terms of 'greener' components for storage systems? What's on the roadmap in storage?

Barrera: There are power efficiencies to be had at the hardware level, just build a more efficient box. Silicon guys are working on those problems. We in the storage industry tend to inherit those technologies.

Deduplication is starting to show up around the industry particularly in archival data. It takes a fairly large amount of computational horsepower, but it's getting cheaper and cheaper, and when you start talking about large amounts of storage, the value for taking that step is going to be there. It'll happen specifically by application types because it turns out that with different applications there are different processes. The use of the intelligence that you have in computers to save the smallest amount of data and commit the least amount of watts when you do operations is going to be an ongoing effort.

IBM is still holding back on adding deduplication to its virtual tape libraries [VTL]. Is there any kind of timeframe there yet?

Barrera: We think the onus on IBM is that by the time we bring a new technology to market, it has to be perfect. Startups get away with more than we can. They perform a great service in the industry as technology trailblazers, but customers buying from us expect nothing unexpected. So it's important to us to make sure things are really ready.

What's the ultimate solution to energy problems in the storage industry? An alternative energy source? Different data recording technologies?

Barrera: First of all, electricity is what all this stuff is going to be based on. There's one discussion to be had around where electricity comes from, and burning fossil fuels better not be the answer for very much longer. Within the data center itself, we've been in this lust for performance and availability of data. We're going to add another lust, and that's going to be energy efficiency, and those things will find their relative importance compared to one another. It will become a new arena for competition between vendors and that's how you will see improvement.

What about replacing spinning drives with solid-state memory?

Barrera: With today's technologies, that's really only possible where the goal is to achieve a certain level of performance if I have a relatively small capacity requirement, like the kind of memory you need in cell phones. Solid-state disk is actually going the wrong way in terms of power utilisation today, but there are things coming, a variety of technologies being talked about in terms of storage-class memories. The idea of those taking over for disk technologies are five and probably more like 10 years away. Disk drives are going to be around for a lot longer yet and so will tape.