Grid Computing: Meeting The Growing IT Demands Of Financial Institutions

Financial organizations are creating opportunities in the form of increasingly complex products. To create sophisticated products in the form of complex derivatives, one requires extensive analysis. One also requires dealing with exposures, interest rates, currency rates, mathematical and statistical models to accurately perform price discovery, risk calculation and estimate market direction. Exotic derivatives like Bermudan swaptions require the use of pricing models like the Brace-Gatarek-Musiela model and simulation using Monte Carlo techniques which require complex and time consuming calculations, putting enormous strain on the IT resources of a financial organization.

To come out with innovative products also requires extensive research of historical market statistics using techniques such as data mining. Data mining, by its nature is a storage intensive process, as it requires storing historical data spanning across various dimensions (time, markets etc.) as well as scales of magnitude.

Regulations such as Basel and Sarbanes-Oxley Act reiterate the need for a financial organization to comprehensively manage its risks, thus creating the need for a computationally intensive computing infrastructure with large storage so that operational, financial and accounting discrepancies can be detected and brought to the notice of the organization as soon as possible.

Nearly all financial organizations need to perform Portfolio Risk Calculations and Counterparty Credit Exposure calculations using various mathematical techniques which are computationally intensive and challenge the current IT infrastructure.

In times of increasing competition, organizations are trying to come up with accurate predictions and estimations in the shortest of times, thus enhancing the need for high throughput applications and platforms to support such applications. In addition, dramatic increase in the volume of trades being carried out through techniques such as Algorithmic trading has led to a demand for low latency infrastructure which can collect market data, analyze it and trigger events based on which orders can be sent to the street, all this happening within fractions of microseconds.

Financial organizations have already started to come up with some solutions to cope with the problems. Some technologies that have been deployed are being further explored to expand the IT capabilities of these organizations. Few of these technologies are Stream Processing, Cluster Computing etc.

Grid computing is a special type of parallel computing which involves complete computers/resources connected to each other in a loosely coupled fashion over a network using conventional network interfaces. It enables resource sharing to produce a large pool of usable resources, such as computing power, storage, network bandwidth, or even software licenses for third party applications and hides the complexity of the underlying architecture from the business user.

This pool of resources can be matched to dynamic business needs. A computing grid i.e. a grid used to harness the collective computing power of a network of computers can provide computing power comparable to that of a supercomputer. Thus sharing of resources is one of the primary characteristics of grid computing. It creates virtual resources which can be shared among a very wide range of users.

Diverse networks & resources make up a grid

The above grid contains trading terminals (Grid users) which harnesses the computing power and storage facilities of Workstations through multiple networks having different topologies. The database server is located in Delhi and file server in New York, but for the users of the Grid it is a single network. Grid Manager performs the functions of resource management, scheduling, and data management and provides security measures for data shared across the grid.

Advantages over conventional technologies

• Resource Utilization: Grid computing, by virtue of its ability to utilize resources efficiently and provide sharing of workload to an extent, can provide a financial organization with the computing power and storage capacity it needs to carry out complex computationally intensive and data intensive tasks.
• Parallel processing: Parallel CPU capacity has for long been exploited in Scientific Research fields. Thanks to distributed computing, parallel processing is finding its way into new fields such as animation, oil exploration and the medical field. Such processing capacity can be utilized by financial organizations for carrying out tasks such as financial modeling, which are time consuming and can be broken down into subtasks which can be processed independently.
• Workload balance in peak hours: During times of peak activity, resources from the pool can be utilized to balance the workload on an entity.
• Virtual resources: Resources such as files and databases can span across systems with various nodes acting as backup nodes with redundant data.
• Software licensing: Tasks to be executed on a proprietary application can be dispatched from other systems to be performed on the system which has the licensed version of the application, thus cutting on the need for multiple licenses.
• Decomposition: A grid can be decomposed into various sub-grids with the users of each sub-grid being governed by a separate set of policies.
• Prioritization: Certain tasks having high priority can be run on systems with greater amount of resources thus helping finish important tasks first.
• Reliability: Reliability using redundant hardware, software and data can be achieved in a Grid environment in a much more cost–effective manner than in a conventional computing environment.

Areas of Application

Grid infrastructure has mainly been applied to applications with high processing requirements and low data requirements. But it is gradually making foray into areas where high data volumes are required. This capability has been extended because of the extension of the definition of a grid to include resources apart from computing power.

The functional areas and processes that can benefit from the utilization of Grid Infrastructure are:

• Front Office: Performs Low latency tasks such as trading and hedging. Generally decomposable into parallel tasks.

• Middle Office: Risk calculation, market scenario simulation and portfolio optimization by their mathematical and statistical nature are computationally heavy tasks. These tasks can benefit from the parallel processing capability of the Grid Infrastructure.

• Back Office: Tasks such as clearing, accounting, and fraud detection are data as well as processing power intensive. These tasks can benefit from the Grid if sufficiently broken into smaller sub-tasks.

• Data Mining: Market analytics based on Data Mining require huge amounts of data storage. These huge volumes are manageable with a Data Grid.

• Financial Research: Organizations bring exotic products to the market after exhaustive modeling and simulations. These applications could benefit from the parallel processing capacity provided by the Grid Infrastructure.

• Risk Management: The risk management systems of a financial organization analyze huge volumes of data on a continuous basis. These applications could benefit from mix of data grid infrastructure and a computing grid.

• Fraud Detection: Fraud detection systems, checking for compliance of trading actions with the internal and external regulatory norms could benefit from the performance boost provided by such an infrastructure.

Challenges

However there are certain challenges which an organization will face while deploying this infrastructure:

• Cost
• File Formats: Existing files being used in the IT processes would be spanning across the whole of the grid infrastructure. Therefore it has to be ensured that the files can be broken up into parts which can be distributed across several locations.
• File Systems: Some existing legacy file systems might be incompatible with the way a Grid handles files which can be resolved by the use of a Virtual/Abstract file
• Heterogeneity: Since a grid can consist of desktops, servers as well as legacy systems, the Grid Management System has to be available for all the software platforms and hardware architectures being used in the organization. Also the tasks being dispatched for processing onto a separate node must be compatible with the processing characteristics and capabilities of that node.
• Resource management: Resource management scenarios include resource discovery, resource inventories, fault isolation, resource provisioning and resource monitoring, which is a very complex task for such a wide infrastructure. This issue can be resolved by combining grid and agent approach.
• Quality of Service (QOS):QoS is a challenge in large scale Grid that comprise of thousands of components from disjoined domains.

Limitations

Like all technologies, Grid also suffers from some limitations:

• Limited Scalability: To enable parallel processing, a task being executed must be breakable into sub-tasks which form independent pieces of execution. If the number of such independent running parts is limited, the scalability of the task being run is automatically limited.
• Trade-offs between Low latency and resource utilization:
• Optimal scheduling: When there is only one resource to be shared for e.g.- CPU , the scheduling can be simple. But when multiple resources such as data storage and network bandwidth come into picture, the complexity of the scheduling algorithms increases exponentially and calls for mathematical analysis
• Access

Grid technology, being a relative newcomer in the financial space, hasn’t had many takers yet. But Financial Service Organizations are gradually waking up to the benefits of leveraging their existing IT infrastructures to get optimal performance, while staying competitively cost-effective. The business drivers in this space namely competition, innovation and information make this technology all the more convincing.

Although an organization has to be wary of the potential drawbacks of such a technology and the challenges in deploying it across its entire infrastructure, it cannot sit back and watch while others reap enormous benefits out of a substantial investment early-on in this space. And as they say, it’s only the beginning…

This article has been authored by Sunny Nagpal from SPJIMR.

Image: FreeDigitalPhotos.net