Peak Performance

Christopher Brown, assistant vice president and director of advanced technologies for The Hartford Life, believes grid computing is a logical concept for IT departments. Devices that participate in the grid can be anywhere that is connected on the network. In [The Hartfords] case, the majority of devices that are engaged are servers sitting in our data center, he says. However, we do have activity under way that is going to be able to utilize desktop machines and servers in other remote data centers. Basically, the grid is virtual in nature. Its really what you bring into it and the devices that are willing to pick up work, process it, and return the results.

What many users find attractive is they dont need to buy additional hardware to roll out a grid solution. What were doing is finding infrastructure were not using all the time, and were putting the grid infrastructure on that hardware, explains Barlow. Were kind of cycle stealing from [the desktops] primary job and setting things up so we can use those assets not just for their primary job but for the secondary jobs when [the computers] are not busy doing any kind of primary work. Driving up utilization always looks good. I know our CTO gets excited when he sees that.

Software Needs

There are multiple resources for grid technology, according to Frank Gillett, principal analyst with Forrester Research. More users I have talked to are using commercial grid software rather than the free stuff, he notes. The free stuff I find more in academia and government.

A company cant just automatically take grid software and put it on its system, Gillett says. You still have to deal with the question of whether your software provider supports it or will be upset to hear youve done this because it is a change in the way
[the providers commercial] software works, he points out. [Integration] varies enormously.

The grid software Nationwide uses from DataSynapse essentially builds the grid, explains Barlow. DataSynapse has got the technology to hook everything together and manage the jobs and workflow, he says. Nationwide currently is focused on identifying packages within Nationwide that could be good candidates to be grid enabled. We have a process I call applications forensics evaluation where you look at an application and ask basic questions to see whether it is a good candidate to put on a grid infrastructure, he adds.
The primary piece of software used by The Hartford is called Condorfree software put out by the University of Wisconsin. It allows these large, parallel jobs to be executed and managed, says Brown. The Hartford uses commodity hardware, networking equipment, and storage. That was one of the aims we had, Brown indicates. To any extent possible we wanted to try to stay away from some of the expensive characteristics of proprietary cluster computing and instead try to design an approach that would allow us to take advantage of the investment we already had made in commodity hardware and software.

A successful use of grid computing means the business problem has to be one that lends itself to the grid approach, such as dividing computation into different resources instead of one big resource. First, its got to be conceptually compatible, says Gillett. Second, the software has to be compatible with that. You take a business problem that lends itself to being broken apart and write software in a way thats [compatible]. Third, assuming you want your [other] software providers to support you, you need their approval. Most providers will say no until you do the standard configuration.

Where They Began

Grid has been around for more than a decade in terms of partitioned jobs running on a number of devices. Its been used heavily in research and in academic circles for at least 10 years. For The Hartford Life, grid computing has been used in a production capacity for about 10 months. Financial services companies were a little earlier on with this [technology], says Brown. Its one of those things that requires the right kind of problem to be solved. It doesnt apply to generic computing. It applies to very certain needs.

Barlow has been assigned as the lead architect on the grid initiatives within Nationwide as the carrier looks to leverage grid computing for its internal applications. Nationwide first got serious about grid computing more than two years ago. We probably started hearing about grid four or five years ago, he says. We looked at a couple of proof of concepts back in 1999 or 2000. The technology really was popular in the academic world, but nobody in the corporate world was looking at it seriously.

The original driver for grid technology at Nationwide came from the Nationwide Financial side, states Barlow. The actuary department, which I consider to be pretty technically advanced, was looking for some alternative solutions because some of its models were running for 30 to 60 days, and it knew that wasnt fast enough, he says. It also knew throwing hardware at the situation wasnt solving the problem.

The actuaries started doing some research in the marketplace, and one of Nationwides vice presidents on the technology side asked Barlow to talk to the actuaries to figure out what they were doing. I got sent to gather information and found, once I looked at it, this was a pretty powerful tool, he says. I realized there was some real opportunity for us to use [grid computing] within Nationwide to shorten the cycle times for some of the information were providing to the business decision-makers.
When The Hartford started its hedging program back in 2003, the carrier realized not only was it a compute-intensive calculation the company needed to run, but the calculation was going to be growing significantly over time. As sales growth went up and time went by, the size of the calculation grew, and we needed to make sure we were in a position to scale effectively, says Brown. In addition, we wanted to take advantage of some of the opportunities grid offers in terms of both good management of large computing jobs and eventually being able to make use of latent computing horsepower weve purchased already but may not be using completely in our environment.

Time Savings

Each actuarial model is different, and completing any model depends on the amount of resources that can be thrown at the problem and how far the job can be broken down. Nationwide has a model it runs called a risk-adjusted return on capital. On a single 2.8 GHz system, it was taking the carrier 36 days to run. Now, we can spread that workload across 50 machines, and it gets done in 26 hours, reports Barlow.

One of the benefits of changing the process for these models is, since analysts can run these models in a much shorter period of time, the time constraint is lifted, and analysts can start running a lot more what-if scenarios. This gives them the ability to run the model 20 times and [produces] a much more accurate view when it comes time to make a decision, says Barlow. It gives them a clearer picture, and they feel more confident in making their decisions.

With its hedging calculation, The Hartford is taking hundreds of thousands of annuity policies, calculating risk out over a 20- to 30-year period fluctuating with market conditions. Prior to enabling our grid technology, this calculation ran overnightabout 10 or 11 hours. Since weve deployed grid and rewritten some software to take advantage of parallel processing abilities, weve reduced the time for that calculation down to approximately 15 minutes, Brown indicates. We run this particular calculation on a daily basis. This allows the carrier to evaluate data in near-real time, which gives analysts more flexibility and opens the door for modeling with even larger data sets and longer time periods.

Among its current uses at The Hartford are actuarial risk management and a hedging program that deals with some of the risk the carrier must manage in variable annuity sales, according to Brown. Thats very compute intensive, he adds.
Brown believes one of the big stories behind the projects success has been a tightly coupled integration between the IT and the actuarial staffs. I dont think we would have made it as far as we have were that not the case, he says. Basically they saw the opportunities for how we could split up jobs and be very thoughtful in the way we used the technology. The IT guys were very engaged in making it work just right with the types of analyses they had to perform. You really have to understand how to make best use of the technology to exploit fully the opportunities that are available.

Few Security Woes

One of the key concerns in grid computing, Brown asserts, is to make sure the grid itself and the management of it are secure. That was part of our core design, he says. With regard to one day expanding our grid to external service providers, that brings in a different set of security issues, such as making sure we have secure interconnects with those folks and making sure we do our due diligence with regard to security assessments with partners as well as encryption of data. The majority of what were doing today is just numbers. There is no personally identifiable information, so from a privacy concern, were not facing those issues.

The new program has been well received at The Hartford Life, comments Brown. Its opening some opportunities in our risk management and hedge management capabilities, he says. Its given [analysts] what they wanted and much faster. As opposed to more conventional, large-scale computing solutions, this actually has saved us a considerable amount of money because we are using commodity hardware to do this.

The carrier is continuing to explore areas where it can make use of grid. Its gone from some of the actuarial stuff into hedging and larger-scale risk management and pricing product development, says Brown. We potentially are looking at it for underwriting support, and other types of work might be good candidates for it. We have a small team actively pursuing that.

Time to Explore

Jim Brackett, leader of the financial technology practice for consultant and software provider Milliman, believes many in the insurance industry sat back and watched the early stages of grid technology development and now are ready to take the next step. I think were starting to see the middle-of-the-road adopters, those who have sat back on the fence, he says. Were starting to see a lot of midsize insurance companies adopt the [grid] solution.

Gillett is cautious about broad statements concerning grid computing. Its not for every application, he says. Most life insurance companies would find some use for it, though, he contends. Any company that is writing Web services or service-oriented software also will find the technology interesting, although lets just say [that may happen] in a couple of years, because right now, I dont think most companies are willing to use Web services software in a grid fashion, says Gillett.

Forrester is in the process of doing research to make the case businesses should be learning about the technology so two years from now they will be prepared to take advantage of it. New applications written in the next five years are likely to be more grid compatible, particularly with companies that produce software for doing their own stuff, says Gillett. The long-term future for grid-type technology is bright. For the near term, there still is a lot of work to do, and its very specific about where you can use it and where you cant and how much new technology you are willing to take on board. This technology looks like a feature that will be added over time to the major infrastructure software vendorsSun, Micro- soft, IBM. Many people in insurance can take advantage of it or should be looking at it.
At this point, life officials at The Hartford have been speaking with their counterparts on the property/casualty side. They are very interested in it, and there are some areas we may be exploring in the reasonable near term, things such as product development and modeling look like a potentially good fit, but were just beginning to explore that, says Brown.

Nationwide is communicating with all its lines of business to look for packages that may be candidates for grid computing. It also is looking at trans-action-intensive applications to see whether grid is a good fit for those types of applications. In the past, people have said grid probably wasnt a good utility to try to use to improve performance of transaction-intensive applications, but DataSynapse has a little different view of that, suggests Barlow. Were looking to see whether we can solve this [transaction] problem using intelligent software instead of just throwing hardware at the problem.

Grid is one of those technologies that is going to take four or five years to mature, Barlow predicts. Its shown a ton of promise so far, and Im pretty sold on it, he says. Its a real technology thats here today that you can use to solve some real issues.