A place on the grid

Technology

technology-gif

A growing number of banks have implemented grid technology for their risk management and derivatives trading businesses, allowing them to borrow spare capacity from dormant computers to process complex tasks in a fraction of the time. By Clive Davidson

The fixed-income group at US bank Wachovia has a risk report task that once took 15 hours to run. By transferring that task to a grid hooked up to existing computers, in particular desktop computers that were often idle, the bank was able to reduce the report run time to 15 minutes. If, instead of using the computers it already owned, the bank had bought new machines to achieve the same performance, it would have needed eight Sun 15K servers, says Robert Ortega, chief architect and head of grid computing at Wachovia. Sun 15K servers cost around $500,000–1 million each.

Canada's Toronto Dominion Securities (TDS), meanwhile, had a derivatives risk report that was taking 45 hours to run on its traditional computing infrastructure. When it transferred the application to a grid, it ran in just two hours. This time could even be reduced further if the bank chose to allocate more resources of the grid to the application, says Barry Childe, director of systems engineering at TDS.

Elsewhere, German private bank Sal. Oppenheim plans to increase the range of structured products it offers to clients. It has introduced Monte Carlo simulation for pricing and risk management, but needs to increase its computing power to do so. The bank compared the cost of buying traditional symmetrical multi-processing (SMP) servers at a cost of around $1 million each with a grid of commodity processors running the Linux operating system. It opted for the grid. Including implementation expenses, the cost was about one fifth of the SMP alternative, says Arno Radermacher, head of IT at Sal. Oppenheim.

So what is this magical technology called grid? Essentially, a grid is a number of computers or central processing units (CPUs) linked together and treated as a single processing utility. Applications implemented on the grid don't run on any one specific machine, the traditional approach to computing. Instead, the application is broken up into various tasks and distributed across a number of computers to be run in parallel. It is an extension of well-established principles of distributed computing and parallel processing, and is in many ways the realisation of the dream of technology visionaries who foresaw computing eventually becoming a resource like water, available on tap.

The grid approach has a number of advantages, particularly for demanding applications such as derivatives pricing and risk management. Just from the examples above, it is apparent that grid can dramatically cut the processing time of computationally intensive applications, can make much better use of existing computing resources (thereby reducing the need to buy more machines), gives organisations more control of how they allocate computing resources and allows financial institutions to do things that they otherwise might not have done because they were too expensive.

The key feature of grid it that is breaks the dependency between application and hardware, says Bob Boettcher, vice-president, financial services, at Ontario-based grid technology supplier Platform Computing, whose Symphony product is used by Sal. Oppenheim, as well as JP Morgan and Société Générale Corporate and Investment Bank. "Grid implements a layer of technology that sits between the applications and the hardware and distributes the workload across what becomes a pool of shared resources," he says.

One of the biggest advantages of breaking the dependency between application and hardware is that it simplifies the task of scaling up the hardware to meet increased demand – either for a shorter running time for an application (for example, pricing or risk reports), or where there is an increased processing load. "The problem with having an application tied to a server is that you cannot dynamically scale – you cannot take advantage of all the other available servers and desktop [machines] to do its processing," says Ortega at Wachovia.

There are several models of grid, but the basic approaches are to either use grid software to link a number of existing computers in an organisation and to use their spare capacity or to create a network, or 'farm', of dedicated machines tailored for distributed processing – or a combination of the two approaches. Like many banks that have been implementing grid, Wachovia has a combination of existing and dedicated machines – around 3,000 of the former, which include dealers' PCs and trading and risk application servers, and around 650 of the latter that are mainly Intel processor-based commodity machines. "We are utilising the resources we have in place already to a much higher degree of efficiency," says Ortega. Sal. Oppenheim will have a grid of 100 dedicated CPUs by the end of the year. TDS would not say how many machines it has in its grid.

Using computers as commodities is another major advantage of grid, says Childe at TDS. Once the dependency between application and hardware is broken, it does not matter what machines are used to process the application, so long as the business requirements in terms of performance, reliability and security are met. So, if a bank wants to run a major risk report within a certain timeframe, the grid technology will seek out the resources necessary to achieve the deadline, either scavenging CPU cycles from idle machines in the organisation, which could include machines in offices in other time zones that are closed overnight, or allocating CPUs in the dedicated grid farm.

"It's all about getting away from silos for each application," says Willy Ross, managing director for Europe, Middle East and Africa at New York-based grid technology supplier DataSynapse, whose GridServer product is used by Wachovia and TDS, as well as Goldman Sachs and Credit Suisse First Boston.

The traditional approach to computing in financial institutions is to run applications or groups of applications on dedicated hardware in a silo for each business unit. This has a number of shortcomings. The provision of processing resources has to meet peak usage, which can mean that for much of the time there is a surfeit of resources. But because the hardware is dedicated to the applications, there can be no sharing of resources across silos – a business unit cannot lend its spare capacity to another that is experiencing a surge in activity. Furthermore, each silo has to have resiliency and back-up and disaster recovery built in, and this drives up costs.

Grid has been evolving over a number of years and many banks have installations that use at least some of the principles of grid. A precursor of grid is what is known as cluster computing, where a number of machines are linked together for the distributed processing of applications, and many banks have implemented these for derivatives pricing and risk reporting.

However, clusters have many of the same problems as traditional computing silos – processing provision must be for peak usage, and they must have back-up and disaster recovery, says Boettcher of Platform Computing. True grid – or what is becoming known as enterprise grid – overcomes the problems of peak provisioning and back-up by the sharing of resources across clusters or silos.

TDS has a global enterprise grid linking its offices around the world, sharing its computer resources. There is no need to replicate the grid at a disaster recovery site because the grid has resiliency and back-up built in, says Childe. This applies on both a small and large scale. If a single computer at a site goes down, the grid software simply transfers the tasks that were running on it to another machine. If an entire site goes down, the grid can move all tasks to another site. While there are some limitations to this flexibility, depending on where the bank runs its core grid processing, there is potential for big savings on business continuity planning using grid, he says.

The principles of unhooking applications and hardware, breaking out of silos and sharing resources is also what makes a grid so easily scalable, and what turns computers into commodities. With an enterprise grid, if a bank wants to scale up its processing power to meet the burgeoning demand for credit derivatives, for instance, rather than having to buy another expensive SMP machine with all its incumbent capital approval process and back-up costs, the bank can simply add some cheap PC-type processors, and only as many as are required for the moment. This process, and using idle capacity on existing machines, can reduce the cost of scaling by 60% or more, says DataSynapse's Ross.

Furthermore, since resiliency is built into the grid as a whole, there is no need to build it into individual machines as is the case with SMP machines, says Childe. Therefore, an organisation can use true commodity processors for the grid – cheap stripped down CPUs with minimal features – and operate a rip-and-replace policy, whereby if an individual CPU fails it is simply removed from the grid, thrown away and replaced with a new one.

However, while it can be fairly straightforward to link an organisation's existing silo-based computers into a grid, the applications written for a single computer architecture generally cannot simply be uprooted and replanted on the grid. An application has to be grid-enabled – in other words, customised so that its processing tasks can be unbundled and distributed around the many CPUs of the grid and run in parallel. Some applications lend themselves more than others to this process, and it is fortuitous for derivatives trading and risk management that pricing and risk analysis are ideal candidates for grid-enablement due to their inherently repetitive nature – for example, the multiple simulations of the Monte Carlo methodology that can run in parallel.

While knowledge and skills in grid-enabling applications are growing in banks, there is also a move among third-party application providers to offer grid versions of their software, with derivatives trading and risk management software suppliers proactive in this area. New York-based front- to back-office systems supplier Summit Systems and North Carolina data management and analytics software supplier SAS have tailored their software for Platform's Symphony, while Toronto-based risk management systems supplier Algorithmics and California-based front- to back-office system supplier Calypso and are among the companies that have grid-enabled their products for DataSynapse's GridServer.

For Wachovia, TDS and others, the move to grid started in derivatives trading and risk management. But now banks are extending the technology to embrace a far wider range of applications. "While the grid was initially used for computationally intensive things like Monte Carlo simulations, we are now pursuing the grid as a comprehensive platform for other applications as well as services," says Ortega. "Instead of having a single consumer asking for a lot of calculations to be performed, which may be parallelised, what you have is many consumers concurrently asking for the same particular function to be performed, and that functionality can be distributed across the grid and conducted in parallel."

The ultimate goal is for the grid to be a central computing resource for an entire bank – a huge pool of resources where business units buy their processing power. And it is not just processing power. Grids manage resources and schedule workloads, and the resources could as well be data storage facilities or software licences. Instead of storing data locally, a business unit can store data on the grid, which will manage the bank's database facilities and allocate capacity as required. The same goes for software licences. Many third-party applications are sold by the licence-per-user, rather than as copies of the software itself. A business unit may in the normal flow of its activities require only a handful of licences for a particular application, but when there is a surge of business it might need two or three times as many. The grid will pull software licences from other business units where they are not being used for critical tasks and reallocate them to the busy business unit.

The key to operating the grid as an enterprise resource, while meeting the various requirements of trading desks, risk management, back office and so on, is the use of service level agreements between the IT department that runs the grid and the individual business units, says Ross at DataSynapse. These service level agreements will reflect the overall priorities and policies of the bank, as well as the individual needs of the business units.

So, for example, a trading desk might specify that when it is using less than 75% of its processing resources, it will lend out 50% of what it is not using, on the condition that should its demand spike it can reclaim the loaned resource within one second. Or another trading desk might require 50 dedicated grid CPUs during market hours from 8am to 5pm, while a back-office service level agreement might specify the processing resources needed (which might be scavenged across time zones) to run an overnight report.

More banks are expected to implement grid technology going forward. In a report published in April, Massachusetts-based research company Tabb Group estimates that current grid technology spending is around $88 million a year, and projects that this market will increase by as much as 1,400% to $1.2 billion by 2010.

Goldman Sachs, for one, agrees there is enormous potential for grid in the financial markets – last November, it led a round of venture financing in DataSynapse, while deploying the company's GridServer technology for its global risk management. However, Larry Tabb, founder of Tabb Group and author of the report, suggests that while it is the smaller visionary vendors such as DataSynapse and Platform that are now developing the new technology, it will be the major data centre management tool vendors such as IBM and Computer Associates that will drive wide-scale implementation.

The next step beyond grid is so-called utility computing. This takes computing to a further level of abstraction, where processing 'power stations' are dotted around the world and connected by a global grid into which organisations can tap for their computational needs in the same way that they now tap into the national grid for their electricity supply. IBM, Hewlett-Packard and Sun are already building these processing power stations, also known as utility data centres or grid centres, with Sun announcing that it will charge $1 per CPU per hour on a pay-per-use basis.

And once the computing utilities are fully in production, the next step will be a commodity market in processing power. Sun and the Archipelago Exchange, the US electronic securities exchange, are already planning to build an online utility computing exchange. Based on the new Sun Grid and Archipelago's electronic matching technology, the new electronic trading environment will allow users to bid on CPU usage.

Being able to bid for processing time in a spot market will give companies enormous flexibility in how they plan and use computing, as well as greater control over the price they pay. What's more, Sun and Archipelago say that once the cash market is up and running, a futures market is likely to follow. It took more than 100 years for the electricity derivatives market to develop; computing as a utility has barely emerged and already there are plans for a futures and options market.

Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.

To access these options, along with all other subscription benefits, please contact info@risk.net or view our subscription options here: http://subscriptions.risk.net/subscribe

You are currently unable to copy this content. Please contact info@risk.net to find out more.

Chartis RiskTech100® 2024

The latest iteration of the Chartis RiskTech100®, a comprehensive independent study of the world’s major players in risk and compliance technology, is acknowledged as the go-to for clear, accurate analysis of the risk technology marketplace. With its…

T+1: complacency before the storm?

This paper, created by WatersTechnology in association with Gresham Technologies, outlines what the move to T+1 (next-day settlement) of broker/dealer-executed trades in the US and Canadian markets means for buy-side and sell-side firms

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here