The TH Interview: Cassatt's Energy Efficient Software for Data Centers
Utility Computing has been with us for a while, in fact since 1961. The idea is to sell 'cycles as a service'; just strap a meter onto a server and then the user pays for whatever computer time they use. It is much like any other utility such as gas, electric, cable, etc.
But is it green? Cassatt, a leading provider of management software for these types of installations, thinks we are headed in this direction... had a chance to blab with Ken Oestreich, director of product marketing, to discuss their product lines for intelligently controlling the power usage of servers in the data center. Such solutions have been proven to cut server power and cooling costs by 50 percent without requiring any changes to existing infrastructure. Here we go.
Cassatt: Hi Mark, good to meet you virtually.
TH: How's it going. Let's start by giving a rundown of what Cassatt does.
Cassatt: We're a software company that's in the "data center management" space, but chiefly in the "efficiency" market. We ensure that large data centers operate at maximum power, capital and operational efficiency; these data centers use something like 1.5% of the total US power consumption... so squeezing efficiency out of them is key. We are based in San Jose, Calif, around 75 people, founded in 2003. The CEO is Bill Coleman (the "B" from BEA Systems).
TH: Does the software require its own separate server?
Cassatt: Yes; it runs on a single server in the data center, and controls all of the others. It's licensed according to how many other servers it controls.
TH: By 'control', what do you mean by that... what can the software do to the other computers?
Cassatt: It can control a number of things.... (1) It can control the "Energy Efficiency" of servers; that is, it will shut-down idle/unused servers, and re-start them when they're needed, much like the IR light switches in conference rooms. (2) It can also "repurpose" servers to do different things e.g. during the day, a server is a webserver, at night it's a financial package. That way, you don't really need 2 servers using up power (and costing money) when 1 will do. (3) Last one - it also does a form of "application availability" which is to say that when a server fails, our software simply finds another one to take its place.
TH: So to control energy efficiency, does it use an algorithm for that or are there parameters that one can set. Or, it is a combination of both?
Cassatt: There are a set of "policies" that the user sets up. For example, you get to define what "idle" or "unused" means. Once the condition is met, we'll then gracefully shut-down that machine. Alternatively, you could have policies that shut down (or partially shut down) groups of servers.
TH: I assume we are talking mostly about energy savings here. Long term, perhaps you get rid of some machines that aren't being used, and perhaps you can change your buying patterns...
Cassatt: There is the energy savings component that equates directly to cash, and there is also the "capacity" issue.... if you can shut down unused machines and divert that power to other machines, that means you don't have to bring more power capacity into data centers. Also, addressing the energy efficiency problem data centers are having is a way in the door -- an on-ramp for us to discuss with customers.
TH: So in terms of electricity, how much savings are we talking about?
Cassatt: It varies by customer, But in a recent pilot we did here in Northern CA, we showed we could save nearly – 26 percent - of the power used by servers alone, not counting what the cooling uses. There are TONS of sources of waste in data centers; for example, servers that are temporarily idle, "failover" servers that sit around using up Watts but waiting for things to fail, and "development" servers that are used sporadically by engineers. Also, doing power management can be the first step toward policy-based management of their infrastructure. This is a great place to start - because it's addressing a specific, immediate data center pain for customers.
TH: Aha. I believe its a safe assumption to assume that the electricity/cooling demands of data centers are increasing i.e. IT staff are starting to be limited by power and cooling constraints, and they want to be more efficient to continue adding applications and servers. Is this true - is the demand for electrons purposed for information technology needs still growing?
Cassatt: That's certainly part of it. I'm not sure what the data center demand growth curve is, but according to an EPA study last year, I think the Watts are expected to double in about 5 years. So it's clearly both a cost and a capacity issue. Plus, if you can make a given data center more efficient, you don' t have to build as many more, and you don't have to add as much electric generation capacity. In fact, we've had the most interest from companies facing serious capacity issues on the power side of things... as in, can't add any more servers; the IT guys end up going, "Now what?" And we work with both IT & facilities to show them an option they hadn't thought of before.
TH: Are we reaching the point where some users will need to do without... there's a oft-mentioned story about how you can no longer run a data center in Manhattan because there's no space left on the rooftop for the cooling. Are we reaching the point of 'haves' and 'have-nots' in the IT world, solely because of the limitations on power/cooling?
Cassatt: There's no more floorspace in Manhattan, either. Absolutely, if traditional operational procedures continue, this will happen.
TH: Ok great. What's the cost of the Cassatt package? Is it one time of continuing?
Cassatt: We charge a flat rate of $200/machine (list price) plus support. That's for Cassatt Active Response, Standard Edition which is the entry-level product that does the power management we've been discussing (turning on/off server based on policy) As you start to control more components of your data center resources, you can move up to our Premium or Data Center editions. (managing virtual machines, networks, etc.)
TH: That number sounds like you are targeting big equipment - $200 would be a tough ROI for a small, mid sized server?
Cassatt: Not really; the average server uses about that amount of electricity in a year - over the 3 year lifetime of a server, the electrical cost to power/cool a server will equate to the HW cost. We priced the product so that there's payback in 1 year or less. There is a Savings calculator from our Website: http://www.cassatt.com/tools/powersavecalc.php
TH: Aha, great. Anything else that you would like to add or think is important to the discussion?
Cassatt: I would like to elaborate on an observation about the industry. So, it appears to me that the industry is focusing on "efficient" equipment (power supplies, cooling, etc.), similar to the obsession over replacing bulbs with compact flourescents, but the industry is missing the boat on efficient operation of that equipment. This is akin to upgrading all of your light bulbs, but then forgetting light switches. So we focus on "efficient operations" in data centers... using equipment more intelligently, only when/where it's needed. We think this is a new model for the industry.
TH: Yes, that brings up a question I wasn't going to ask but I will now. You are talking about the efficiency of the production costs and the system costs; as systems get more efficient, there is a tendency for use to actually increase instead of decrease. In the data center case, software packages like Cassatt's free up a ton of surplus electricity at the system level. Obviously, what many companies are going to do is take that surplus energy and run more servers with it -- this will essentially plateau them off of their current level of consumption. And at 1.5 percent of the grid's electricity, we are already using a ton of energy for IT-related business. What's your opinion here - is this too much electricity, will we see a world where 10% of its electricity goes into computing, or it consumption gets reduced from its current state?
Cassatt: This current growth curve is not sustainable. I think we'll see 2 trends... one is a move toward more operationally-efficient compute equipment and data center management, and the second is a move towards more centralized computing, this is not unlike how electricity is generated today. Very large, automated and efficient "data center service providers".
TH: like... Google. I have often suggested that everyone should be on Gmail as the most efficient way to do that service!
Cassatt: Yes, but perhaps in a more generic sense, like Amazon, and the existing computing service providers (Equinix, Savvis, Sungard) will play too; it's sometimes called "utility computing". Bill Coleman expects a convergence of providers - the "Googlrizons" of the world - with service levels of the applications being the most important thing, and the underlying infrastructure being something that is commoditized.
TH: I think that is very interesting - Googlrizons. Of course, that will increase demand for the services, as the price will fall even farther. However, I do think that that seems like a fairly realistic future.
Cassatt: Yep. Here's a 15 min keynote Bill gave on the topic at the Velocity conference: http://blip.tv/file/1018376
TH: Great, anything else?
Cassatt: Another tidbit: usually the first thing we hear from customers: "Isn't it bad to turn off servers?" We put some useful stats together & address that and other myths here: http://www.cassatt.com/myths.htm
TH: Terrific, thanks for your time.
Cassatt: Good chatting with you, Mark.
Designing Radically Efficient and Profitable Data Centers
The Problem Is Us - U.S. Data Center Growth Spawning Climate Disaster
US Data Centers May Be Forced To Green Their Operations: They're Running Out Of Space And Looking For A Reason To Expand