An Overview of Grid Computing in Financial Services

Introduction

The financial services industry today is one that thrives on information. By and large, the services we provide to our clients incorporates the provision of information as a major portion of the value added. This information goes beyond the latest market prices and incorporates portfolio rebalancing scenarios, risk modeling/analysis/management, and market analysis. As our models and our analytical methods become more sophisticated, we are hitting a wall: our model-generation capabilities are surpassing our ability to analyze data based on those models. There's simply too much data and too many analytical factors; we've been forced to compromise our models' accuracy in order to bring computation times down to reasonable limits.

Grid computing promises to break down that wall for the financial services industry.

By pooling computing resources within an organization or across organizations, computational power of a new order of magnitude is now possible. Not only will this allow us to finally realize the promise of our sophisticated models and terabytes of data, grid computing will also give us the capability to find information in mountains of data and deliver that information to those who need it most.

Definition of Grid computing

Before we get ahead of ourselves, let's define what we mean by "grid computing." Grid computing is a form of distributed computing; the pioneers in this field are Dr. Ian Foster of Argonne National Laboratory and the University of Chicago, and Dr. Carl Kesselman of the Information Sciences Institute and the University of Southern California. In their 1998 seminal work, The Grid: Blueprint for a New Computing Infrastructure, they defined grid computing as follows:

"A computational grid is a hardware and software infrastructure that provides dependable, consistent, pervasive, and inexpensive access to high-end computational capabilities."1

Unfortunately that definition is a bit vague to get a true understanding of grid computing today. What we're talking about is a distributed-processing technique that spreads a computation among multiple machines.2 This different from other distributed computing techniques like Massively Parallel Processing because the processors are loosely coupled (they are not a part of the same physical computer). Still, the idea is not fundamentally different: it is essentially the "ability to farm out calculations from a single application over multiple, physically separated CPUs."2

Increases in processing power are made possible by breaking down a specific problem into many smaller pieces and distributing them to available computers. This is why grid computing is particularly well suited to numerical problems. For example, BankOne uses grid computing to break down the mathematics of derivatives trading into smaller pieces, distribute those pieces among the grid, and aggregate the individual answers to achieve a whole result.2, 3

Such capability has existed in academia for years and only recently has come to the forefront of corporate consciousness. Driving the new corporate interest in grid computing is a sluggish economy in which companies are striving to achieve every possible efficiency. Take into account that only 10-20% of the typical corporate desktop computer's CPU cycles are actually used. Add to that the fact that, "one could argue that the average company has a few terahertz of computing power."4 That combination of capacity and utilization would be completely unacceptable for most corporate assets like manufacturing plants. For an industry whose main product is information and whose factories are computers, why should a 10-20% level of utilization be acceptable? Grid computing offers the technology to dramatically reduce that under utilization by apportioning computational tasks on an on-demand and by-availability basis.5

Uses of Grid Computing in Financial Services

Given an admittedly rough definition of grid computing, perhaps you are beginning to see the potential of this technology for the financial services industry. Indeed, many of those in the vanguard of financial information systems already take advantage of grid computing.

Data synapse, a New York City supplier of grid-enabled middleware, boasts a laundry list of over 60 clients in the financial services industry. Examples of Wall Street firms using the products of Data Synapse and similar vendors are BankOne, Charles Schwab, Wachovia, JP Morgan Chase, Deutsche Bank Group, Bank of America, Abbey National Group, and Pacific Life Insurance. The diversification among these companies -- banking, investments, insurance, etc. -- indicates the widespread popularity and application of grid computing in financial services.2,4,6

By considering the computational tasks these diverse companies have in common, we can gain an insight into the major opportunities for grid computing. The most common use is in risk analysis and modeling. Risk is present in every financial transaction and the risk/reward concept is a cornerstone of the industry. Therefore, a competitive advantage can be gained by the company that has the most comprehensive and deepest insight into the potential risks and rewards for any given transaction.

Even with grid computing, a truly comprehensive insight into the risk of a transaction is not possible; there are unknown quantities in the equation. That means that we must create models, or estimates based on known factors. This is something that, by necessity, humans have become quite adept at. However, until the arrival of grid computing, our ability to process information using these models has trailed our desire for analytics. Grid computing, by making available a vastly larger pool of computing resources, can improve the accuracy of our risk modeling in two ways. First, it can incorporate more factors in the risk model equation. Second, it can process much more data with that equation, producing a result that carries much more confidence. 7,6

A similar use is portfolio rebalancing. Ideally, a portfolio rebalancing program would take into account all of the goals of the client, the universe of all available financial instruments, and all known data regarding the history of each instrument in that universe. Clearly this is an impossibly immense amount of data to process (think just of the entire pricing history of every common stock issue in the United States). Thus, we must estimate. Again, grid computing offers the ability to incorporate more factors and more data into the computation of our estimates. At Charles Schwab, grid computing allows representatives to run portfolio analysis scenarios in fifteen seconds rather than the previous four minutes, a 94% reduction in time required for computation.2

The time required to calculate the results of a model is critical in financial services. In Schwab's case, it just meant having the ability to discuss various scenarios with customers and improve service. However, for the quantitative investor, modeled predictions dictate trading decisions. For a volatile stock or bond, a decision based on four minute old data may no longer be valid. A prediction calculated within fifteen seconds, though, may be quick enough to make real-time trading decisions based upon truly current data. This could revolutionize trading through the creation of automated quantitative trading systems that make instant and confident decisions based on current data. In the words of one investment professional, "we are booking larger, more exotic, and more lucrative trades with more accurate risk taking."7

While there are clear front-office business advantages to grid computing, back-office and especially IT administration can also benefit.

An interesting area in which grid computing can improve efficiencies in corporate computing resources is digital content distribution. It's great that grid computing enables us to produce greater volumes of information, but that information is only useful once it gets into the hands of those who need it. While the current focus on grid computing regards pooling processing resources, that is not the extent of the technology's capability. The same technology can, and has been, harnessed to pool networking and data storage resources. Grid Delivery Server and Delivery Management System by Kontiki provide such a distribution service. They deliver video and large files to users on demand by determining the most efficient way to deliver the file over the network. They can store large volumes of data across distributed storage devices. This means that large information stores can be located closer to the users, reducing the need to utilize WAN connections to headquarters or centralized data centers.4

Because grids provide on-demand computing power to applications through the pooling of computing resources, they provide two major ancillary benefits to IT administration. First, the utilization of the corporate computing assets goes from 10-20% to potentially 90% or higher. Second, because the resources are pooled, the application using the resources is at too high a level of abstraction to know or care about the details of the underlying hardware. That means that when more computing power is needed, inexpensive resources like Intel based servers running open source operating systems can be added on an as-needed basis. Similarly, storage resources can be added by adding disk space on an existing Storage Area Network (SAN) or adding hard drives to existing PCs connected to the grid. The consequences of such easily predicted, horizontal, platform-independent scaling of computing power are greatly reduced hardware procurement costs.2

Risks of Grid Computing

While grid computing has many advantages and a number of specific uses within the financial services industry, it is certainly not without risk.

First, and probably most obviously, distributing processing among a very large network of heterogeneous computing devices represents a very complex task. The software to manage these resources as node availability fluctuates and to break down the processing tasks must be very sophisticated. This pooling of unlike resources also presents a sizable management issue for IT to handle. Because each node in a grid is now an important part of the entire enterprise computing platform, inventories must be more tightly controlled and systems must be better maintained.2

This complex architecture also re-raises questions that came with mainframe host computing thirty years ago: how do we account for the processing time, hardware and maintenance costs? No longer will we be able to charge departments for the cost of another server that their software required…this is the downside to the advantage of horizontal scalability. How will we manage security? Authentication issues that exist now are multiplied by the number of nodes on the grid! How will we schedule jobs that are contending for the pool of resources? A "meta-scheduler" is required to schedule time on computers which must each schedule their own jobs; it must then manage the aggregation of piecemeal results from each node. There is an issue now that not only must the server administrator specify priority of contending jobs, but those with control over the individual nodes in the grid must specify the priority of grid jobs vis-ˆ-vis their local computations.5

A similar but perhaps more subtle risk of grid computing is the need for a new licensing structure for COTS (commercial off-the-shelf) software. No longer does it make sense to talk about licensing in terms of "per CPU". With a grid, the computing resource pool is constantly shifting -- the number of CPUs is a function of the availability of the computers that make up the grid. Additionally, some applications that take advantage of grid computing will consider processors of secondary importance, in favor of other resources like network bandwidth and storage space. The difficulty of this licensing dilemma reaches a new order of magnitude when you consider the inclusion of computers across multiple organizations in the same grid.3

Finally, making a move to grid computing represents a cultural shift both for IT and its clients within an organization. For IT, the administrators must tackle a new, complex administrative task. The developers must adopt a new programming paradigm. Managers have new vendor relationship issues to consider or must manage the use of open source contributions like Globus. Clients must recognize the capabilities and limitations of grid computing. They must also adhere to the new accounting measures that are put in place to pay for the miracle of the grid.

Conclusion

From prepackaged industry-based solutions from big players like IBM to small niche startups like Entropia and open source projects like Globus, Grid computing technologies are entering the mainstream of computing in the financial services industry. It is, slowly but surely, changing the face of computing in our industry.

Just as the internet provided an means for explosive growth in information sharing, grid computing provides an infrastructure leading to explosive growth in the sharing of computational resources. This is making possible functionality that was previously unimaginable -- near real time portfolio rebalancing scenario analysis; risk analysis models with seemingly limitless complexity; and content distribution with speed and efficiency hereunto unparalleled.

In addition, grid computing provides a means for organizations to scale their computing capacity in a much more efficient manner. Rather than having to buy and configure new servers for each new application, a grid administrator can incrementally add resources as needed -- extra disk space here, a couple of extra processors there. Besides being much more financially efficient and fiscally responsible, such an infrastructure reduces a lot of the complexity involved in managing enterprise computing resources.

However, grid computing can be a double-edged sword. As much as it eases some management tasks, it raises new challenges. Grid computing brings up new questions with regard to accounting for resources and charging the costs back to clients. Apportioning shared resources may be a politically difficult task. There's also a new problem of providing high availability and predicting process run times given the ebb and flow of the availability of underlying grid nodes.

As with all new technologies, grid computing is not a panacea for information technology within the financial services sector. However, used for the right applications and with sufficient planning and forethought, grid computing will become an indispensable part of an organization's technology portfolio.

References

1. Foster, Ian and Carl Kesselman ed.: The Grid: Blueprint for a New Computing Infrastructure; Morgan Kaufmann; San Francisco; 1999.
2. Schmerken, Ivy: "Girding for Grid," Wall Street & Technology; New York; April, 2003.
3. Greenemeier, Larry: "Schwab Girds for Grid Computing," Information Week; Manhasset; February 3, 2003.
4. Marsan, Carolyn Duffy: "Grid Vendors Target Corporate Applications," Network World; Framingham; January 27, 2003.
5. Franklin, Curtis Jr.: "Grid-dy Determination," Network World; Framingham;
6. Bielski, Lauren: "Got Grid?," ABA Banking Journal; New York; December, 2002.
7. Bills, Steve: "Wachovia, Morgan Chase Bullish on Grids," American Banker; New York; Sep 11, 2002.

If you have any questions or would like to contact me for any reason, please email me at j.eckles@computer.org.