|
From Resource,
February 2005
Copyright by LOMA
Grid Computing At Hartford Life
Find out how Hartford Life is using grid computing to help tackle its complex computational problems.
By Tammy J. McInturff
Grid computing, which is already being used in some academic and research communities, is making its way to the life insurance industry. Essentially grid computing involves sharing computing, data, storage, application, or network resources, to ultimately allow companies to solve large-scale, complex computational problems.
Hartford Life is among the first life insurance companies to implement grid computing. Resource recently talked with Vic Severino, senior vice president and CIO, Hartford Life, about how The Hartford is using grid computing to help manage the risk for income protection benefits associated with its variable annuities.
Resource: Why did The Hartford decide to implement grid computing technology?
Severino: It really was out of pure necessity. We have some complex products that require a high level of computing power. Essentially we had a very pressing business need—and I would say this was probably one of our most important business initiatives—to essentially manage the risk as a result of these complex products. We needed a lot of accessible and stable computing power. So we did some research and talked to some of our investment banking partners; the investment banks have been using grid computing for a few years. We decided to embrace grid computing for ourselves. What got us into it was just the fact that we had a very pressing business need.
Resource: How did you complete this project—on your own or with a vendors help?
Severino: We completed this project very much on our own. As a matter of fact it is such a new technology in the insurance industry, that others were calling us for assistance on how to do it. So it was interesting because we were breaking new ground and vendors really couldn’t help us. We eventually chose grid computing software from the University of Wisconsin called Condor; it is open source software. We chose the Condor software because it is one of the oldest grid computing software tools around; so it is mature. We have a tremendous amount of confidence in the Condor software.
Resource: What is a computing grid? Explain how it works.
Severino: Grid computing software allows you to access a lot of computing power in a way that a makes that computing power very accessible. If there is a problem with any one server it is reliable enough to recover from one server being out—if a server goes down, then it actually pushes off the work that would have been on that server to another server. For example, picture dozens, if not hundreds, of servers or desktops that are strung together in a big grid almost like a spider web of nodes of computing horsepower. What grid computing software lets you do is access all those servers almost as though they were really one. And it allows for scalability so that you can add servers very easily without a lot of reconfiguration. It is also very stable.
Resource: How is The Hartford using grid computing technology?
Severino: Essentially we are using it for risk management associated with our variable annuities. One of the features of a variable annuity is called a living benefit and basically what it allows a policyholder to do is to guarantee that they will be able to at least get their principal back after investing in the annuity, assuming that they withdraw their principal over time. So that guarantee is essentially a guarantee on the stock market. So you could invest in our annuity and no matter what the stock market does if you have invested a $100,000 you will get your $100,000 back over time. What we need to do is hedge that risk, in other words, make sure that no matter what happens in the stock market we have invested in such a way that we will earn enough money to pay back that principal. That business problem requires a lot of computation because essentially what we are trying to do is simulate market behavior and policyholder behavior. So basically what we do is create market scenarios and policyholder behavior scenarios which require a huge amount of computing horsepower.
So we simulate what the market is going to do over 15-20 years and what policyholders are going to do with our products over that period of time. Then we calculate the risk that those guarantees bring us. And when we calculate the risk we can figure out how to hedge that risk using other instruments that essentially will be moving in the opposite direction as the market. So if the market goes down these other instruments go up in value and allow us to pay back the principal over time. But all that analysis requires a tremendous amount of computing power.
Grid computing allows us to access servers and desktops simultaneously so that we can harness that power in a way that makes it easy for us, that doesn’t force us to engineer something so complex that it could be unstable. So in short we are using grid computing technology to help manage the risk for income protection benefits associated with our variable annuities.
Resource: Has it improved your risk management capabilities?
Severino: Not only has it improved our risk management capabilities, I think it actually has helped us to build a very sophisticated risk management capability. Without grid computing we probably wouldn’t have had the computing horsepower available to do what we wanted to do.
Resource: What do you expect to gain from grid computing? What are your main goals?
Severino: Well number one was scalability. As time goes on our risk management modeling gets more and more complex and requires more horsepower. So we wanted to be able to scale to that and bring to bear more computing horsepower to do the ever more complex calculations.
Second, we obviously wanted scalability with stability. As we brought more servers and desktops onto the grid we didn’t make it any less stable by having a bigger environment.
The third goal was cost savings. One of the most technologically interesting things that grid computing lets you do is utilize unused computing power from desktop PCs, print servers and file servers. Even during the day when you are using your PC it is virtually unused—a very small percentage of its capacity is utilized. Most of the time the average desktop PC sits idle even when you are writing e-mails or word processing, what grid software lets you do is actually harness that 80-90 percent of available CPU even if you are using the CPU for other things.
We have done tests that show that we can take dedicated servers and non-dedicated PCs and together create a grid, but this isn’t in production yet. Obviously if we could take advantage of desktop computing power that we have already purchased or file and print servers that sit completely idle at night and bring them into the grid then we would not have to buy as many dedicated machines.
However, we are not only saving money on hardware and software development, we are also saving on IT maintenance costs for two reasons. First we are saving because we have less equipment than we would have had to purchase and fewer people that are needed to support that equipment. Second is because the environment is so stable that there are fewer application support people. So it saves us on infrastructure support and application support.
Cost savings is a very real part of the equation but because we were really solving a very specific and important business problem, it was not the number one priority for us. Thus far we have saved probably about a million dollars, just in the software and hardware purchases that we have been able to avoid because of grid computing. It is really difficult to tell how much we are going to save because the complexity of our analytics grows by such an amount that I don’t know two years from now how many servers I would have had to purchase if I didn’t have grid computing. But it will certainly allow us millions of dollars of savings.
Resource: Can you clarify the difference between clustered servers and grid computing?
Severino: A cluster of servers essentially behaves as though it was one big server. Servers in a grid behave independently, though they often run in collaboration. Clusters rely on homeogenous, often proprietary, hardware, while grids are generally built from a variety of devices and configurations. In fact, many large grids employ clusters as nodes within the grid.
Also, you typically have a maximum number of servers that you would put in a cluster. A grid allows you to put almost an infinite number of servers together. For example there is an application to look for a cure for cancer that was developed by Oxford University. Essentially you can plug your desktop into this massive grid that they are using to do biological modeling to help them find a cure for cancer. At any one point and time I think there are 250,000 PCs working for this public grid. We used that very application to present to our senior executives, to try to give them a sense of what grid computing is. With this application we were able to show them the true power of grid computing.
There is no clustered environment that has 250,000 servers or desktops attached to it. So the grid gives you a lot more flexibility in terms of adding CPU power. Grid computing is best utilized with a problem that can be sort of pieced up into 1,000 or 10,000 ways and then eventually brought back together; it really requires problems that can be
divided up.
Resource: What lessons were learned during the process of building the grid?
Severino: The major lesson that we learned was that grid computing is a viable technology. It is a huge technology and it was surprising how good the technology really is. Grid computing seems to be a technology that has been kind of hidden behind the scenes. It is a very stable, mature technology that for some reason just never made it from academia to the insurance industry.
The other lesson we learned is that even though it is a very viable and mature technology, it still takes a lot of engineering prowess to bring it to a point of adding business value. It is a different way of thinking. It breaks some cultural bounds; we are having some issues in getting people to let go of their CPU power which is why we are using dedicated servers right now. But eventually we will be using the spare CPU from non-dedicated desktop PCs.
I think another lesson learned is that there isn’t a lot of knowledge or help out there because it is a pretty new technology in the corporate world. There are some products that are coming into the public realm, in other words they are not based upon open source software. But most of those are just starting and a lot of those vendors are really cutting their own teeth on the technology. So there isn’t a tremendous amount of help to be had on this technology right now. The situation will be different three or four years from now.
Another lesson was that there are several different definitions for grid computing. Some vendors may have grid management products but that isn’t necessarily the same as grid computing. This unfortunately tends to confuse the issue. There are other companies that are talking about utility computing, which has kind of gotten a bad name. But grid computing is not the same as utility computing. However sometimes you will see grid computing used almost as a synonym for utility computing, which is clouding the issue. So we have to be careful too about what the definition of grid computing is.
Resource: What is the best feature of grid computing?
Severino: The best feature is really its scalability. I was definitely losing sleep at night in terms of scalability when I saw what these guys wanted to do as far as analytical complexity. I would sort of draw a trend chart of how many servers I would have to bring into a cluster to be able to solve these complex problems and without grid computing I couldn’t do that in a way that I would feel comfortable.
The grid can run across any type of networked computers, desktops, file servers, print servers, at the same time. It can also run on any operating system regardless if it is Windows, Linux, UNIX, etc. And we could run multiple operating systems at the same time. So in other words some servers could be on Linux and other servers could be on UNIX and we can attach them to the grid. The grid sort of makes that irrelevant so we could put anything with a CPU in it, with almost any operating system. It has a tremendous amount of flexibility.
Resource: What are the security concerns or issues with grid software?
Severino: The main security concern is ensuring that the grid does not introduce any security holes on someone’s desktop and at this point in our testing it doesn’t seem to present that type of problem.
Resource: How have your producers, senior management, IT team, etc. responded to the project?
Severino: They have responded very favorably. We won an award, called our President’s Award, which is our top award in the life company and we are nominated for a Chairman’s Award, which is an award that is across the entire Hartford. So it has been very well received. Our president really likes the concept. When people see it and experience it they really see the benefit and power of it. So it has actually captured senior management’s imagination. It is a technology that almost anyone could understand once they see it, whereas if you talk about customer relationship management or other types of technology it might be difficult to really envision the power of it.
Resource: What is the downside to grid deployment?
Severino: I think the biggest downside with grid really is that you can’t solve every business problem with a grid infrastructure. It really is most amenable to problems that can be divided up into smaller problems, then farmed out to the grid and brought back together again with the answer. So for a lot of transaction processing applications the grid doesn’t really help. I think the grid environment really is best
suited and is targeted for heavy computational problems. So if there is any downside, it is that you can’t throw it at just any business problem. There is a specific category of problems that it solves.
Resource: Where are you today with this project?
Severino: It has been in production for about five months. We went through a lot of testing before going into production. We have also built a prototype that will be able to take these analytics, this sort of modeling platform, and run it on anything whether it is a desktop or a server and virtually any operating system. That isn’t yet in production but the prototype has been built and tested. We have done a production sized test and it actually runs faster than production today which was another surprise.
We are rolling out more grid environments. We have started to put together a grid environment for product development and we are also going to put a grid environment together for another area that manages risks at a more corporate level. So the grid is starting to sort of propagate its way through the company.
Resource: What problems or setbacks did you encounter?
Severino: I think the biggest problems were at the beginning, trying to choose which grid software to use and the setbacks were just in getting started because there isn’t a lot of expertise. One of the reasons we chose Condor, the University of Wisconsin application, is because they were very amenable to helping us. It is a slow ramp up time because you can’t just call an expert in to help in getting real business problems solved using grid computing.
Resource: What are your future predictions for this project?
Severino: My future prediction is that a year from now our grid will be four times the size it is now. It is just increasing geometrically. I always think I know how big it is going to be and then I am always surprised at how big it is really getting. It just keeps growing because the more computing power you can bring to a problem the more you see how beneficial it could be.
Resource: What advice would you give another company going through the same process?
Severino: Well my advice is if you have complex computational problems that can be divided up, to definitely look into grid, because it is an affordable and mature technology. However, since there isn’t a lot of expertise in building applications for it yet, a lot of the expertise will have to be homegrown and you’ve got to get the right people and the right engineering acumen to ultimately be owners of it. Grid computing is still new, especially in the insurance industry, and while it is very viable as a technology there isn’t a lot of external help, but that will probably change in the next two or three years.
|
