Silver Lining?

Unwilling to be left behind, ambitious investment firms and exchanges are looking for ways to prepare for the recovery with cloud computing. By Bob Giffords

The middle of a global recession may not seem the ideal time to consider new technology, but for Peter Kelso, CIO of DB Advisors, at Deutsche Asset Management, this is no ordinary crisis. "Some of the old ways are gone forever and new challenges keep coming," he says. "We have to rethink our assumptions about staff, about systems and the way we do business."

One of the new challenges that has appeared in the past few months is a technology concept that is raising interest and concern in financial services. "Cloud computing, for example, has a compelling cost argument. It is something seriously worth looking at, even if there are still some challenges," says Kelso.

Regis Moneret, CTO at investment firm Reech AiM, agrees. "Cloud computing is not a new concept, but more an extension of Web 2.0," he says, adding that the improved technology model for pay-as-you-go is "very attractive."

They are not alone in their enthusiasm. Although some sell-side investment firms remain to be convinced that the cost case for cloud computing is there yet, interest in it is spreading. In an independent survey into current IT trends and challenges, undertaken by Vanson Bourne on behalf of IT infrastructure services provider Savvis, 64 percent of financial services IT professionals in the UK, US and Singapore agreed that cloud computing will be a priority for them in the future.

Despite this approval, the challenges persist. "Many banks are actively evaluating clouds, but few are using it in earnest," says Dave Malik, director of solutions architecture in the advanced services team at Cisco Systems.

It is important to distinguish between public and private clouds, says Roji Oommen, director of financial markets at Savvis. "Some institutions may shy away from considering public clouds for certain applications, but private clouds can help deliver similar economies without the security concerns. In general, we see widespread interest in leveraging managed cloud computing services among banks," he says.

For Christopher Clack, director of financial computing at University College London (UCL), we are moving toward a cloud-bursting model, whereby most processing will be done in an internal cloud, from which we can then burst out into a trusted external cloud on demand. "This might allow us to support a lower capacity internally that can be topped up when needed," he says. "Less time-critical jobs could be cloud-bursted, where latency is important for the remaining internal jobs."

Unlike other emerging technologies that usually debut and take hold in the US or the UK first, this time Asia may be taking the lead. Firms in Singapore realize that being small means that they must move faster to survive, says Songnian Zhou, CEO at Platform Computing. "They are very ambitious and they want to create the tools to let local business flourish," he says. Platform Computing is forming partnerships with the Infocomm Development Authority in Singapore to provide technology infrastructures to support a national cloud utility for local business and the country's global customers. "We have also set up a cloud innovation center to assist in the training and development of cloud-enabled applications and proof-of-concept exercises," he says. "There is a real momentum building as people see how they can both save money and still expand their business."

Challenges large and small

The main challenges facing early adopters are likely to focus on security, application entanglement-hard-to-move pieces of a tightly interconnected infrastructure-service level agreements (SLAs), performance, regulatory restrictions, and economics, according to Evangelos Kotsovinos, vice president of hosting at Morgan Stanley. "It needs to be demonstrably cheaper," he adds. Nevertheless, the firm is moving ahead.

"We leverage the internal cloud concept in a number of areas," says Kotsovinos. "For example, we offer our internal customers the ability to deploy applications, common services and middleware components on-demand on a shared, consolidated infrastructure. This provides efficiency and has good acceptance, as users are not required to perform heavy lifting on the bare metal. It also improves time to market and reduces costs."

For Kotsovinos, cloud computing is a new business model more than a new technology. "I think of it as an insurance policy against capacity risks," he says, "Customers of cloud computing offload the risk that they will run out of capacity, or that they will have excess capacity, to a third party, internal or external. This leaves the providers of cloud offerings with potentially large aggregate capacity risks."

Some smaller firms see the challenge more in terms of legacy mindset. "Cloud computing allows us to outsource extremely low-volume, high-value services," says Simon Hazlitt, co-founder of Majedie Asset Management. "Most IT people think only in terms of Henry Ford's concepts of high-volume, low-value, commodity operations. Scale is no longer a necessary precondition for efficient margins."

Hazlitt explains that his firm has never had an IT person working for Majedie Asset Management since the firm was formed in 2002. Instead, they buy solutions from vendors such as Bloomberg, Omgeo, Microsoft or salesforce.com. "We see cloud computing as the platform for software-as-a-service. It is what we have always been doing," says Hazlitt. "We call it direct computing. It's immeasurably cheaper so we can run the organization with just 19 people."

Indeed, Hazlitt argues that with fund management, the bigger you are, the harder it is to do things well. "We have closed our doors to new business with £3.5 billion ($5.4 billion) under management," he says. "We share our economies with our clients. We have shown that smaller can be cheaper and better. Companies have to move beyond the 20th century paradigm for technology governance."

Reech AiM has adopted a different approach. "We have built our own cloud internally based on a server grid of around 100 nodes," says Moneret. "The architecture is in place, it works, and now we are in discussion with other hedge funds to agree a suitable SLA for sharing it with them later this year."

 Understanding how to optimize investment applications for the cloud is key. Monerat notes that clouds can get blocked easily when handling large data volumes in ways that may not be immediately obvious. "Managing peak loads can be difficult. You have to get the right monitoring and grid management tools in place to make it work," he says.

Mike Stolz, vice president of architecture at data-grid specialist GemStone Systems, believes that cloud adoption could be driven by smaller investment firms once data vendors like Bloomberg or Reuters get plumbed into the clouds with pay-per-use licensing models. "Hedge funds rely mostly on public market data anyway, so the security issues are far less serious," he says. "The private data can be held back, with only the information pertinent to the computation pushed into the cloud to be used in conjunction with the public data. The data fabric would then ensure functions execute where the data is, drastically reducing data transfer volumes." 

Stolz stresses that cloud computing creates new data management challenges that require new solutions. "Our GemFire data fabric is now able to co-locate processing with data, and distribute either data or function calls to the appropriate server at the right time, while dynamically growing or shrinking the server pool on demand. We call it the elastic data fabric. This enables efficient, extreme transaction processing while optimizing costs in a pay-per-use economy." 

Even if data fabrics can help new applications, legacy systems still need support. "Clearly, banks will need to shift terabytes of data," says Oommen at Savvis. "So it is very important to have a large-scale utility storage capability or dedicated storage area networks (SANs). Savvis will also support different middleware stacks for whatever data management framework the client may choose." 

Work in Progress

While the cloud computing forecast may be bright, the journey will be long. "Cloud is an ambition, but the technology is still evolving," says Malik. "On Cisco's cloud roadmap, workload mobility and multi-tenanted, pay-as-you-go, virtualized servers, are key components that are enabled by our Unified Computing System (UCS)." He adds that as applications move dynamically, ubiquitous virtualization requires policies, network properties, storage attributes and security profiles to move with them and adapt automatically keeping the same original configuration, with full visibility and audit trails.

Clack at UCL agrees that software can still be very difficult to move on demand, especially if it includes hard-coded details that are specific to hardware and/or an operating system. "Making jobs run in parallel to gain speed is notoriously difficult," says Clack.  "Greater abstraction in the language may be key to overcoming some of these difficulties, for example, using a functional language such as F#-but clearly there is a huge inventory of existing code that we have to deal with," he says.

"Applications behave differently in terms of latency, memory and data," argues Clack. "There is a lot of work involved just to monitor and calibrate demand. SLAs may have to be two-way agreements-what the job expects from the cloud, and what the cloud expects from the job. How will clouds treat jobs that ‘go rogue' and consume too many resources?"

The devil is in the details. "For cloud-style dynamic allocation to take off, accounting and billing will have to be fully automated and allow for significant flexibility in terms of charging units," says Kotsovinos. "This is a much harder problem than it originally appears to be; consider that depending on the application, customers will want to be charged, for example, by the year, hour, number of messages per second, gigabytes, or a flat fee."

As for security, that may just take time. "We trust people to keep our data safe," says Hazlitt at Majedie. "If salesforce.com lost data, they wouldn't have a business, so we use them at a fraction of the cost of an in-house application. We don't see a problem with security for clouds. Vendors just must provide it."

 Latency is another challenge that must be addressed. "It is a key factor for some distributed applications," says Oommen at Savvis, "It is important to consider locality within the dynamic provisioning policies. Where we have local datacenters and fiber connectivity between them, latency in general should not be an issue." 

Of course, for applications to move to the cloud without facing an adverse latency impact, clouds will need to be globally distributed. "You can't beat the speed of light yet," admits Kotsovinos.

Clouds in use

When it comes to cloud computing, which low-hanging fruit are ripe for harvest? Software development and testing services, according to Savvis's Oommen. "It doesn't have the same security issues as production; it exhibits a highly variable demand profile and it is typically one of the first budgets to be cut. Credit, counterparty, and portfolio risk systems are also attracting attention because of the cost savings potential caused by sudden surges in demand," he says,

Stolz at GemStone agrees: "For larger investment banks, our data and function in motion concepts are allowing complex risk analytics and aggregations by counterparty, sector or collateral groups to be carried out in real-time along with trading." Queries can be added on demand and are automatically scheduled as part of the transaction flows. "Once again the fabric itself will worry about elasticity of the data grid based on servers being provisioned," he says. "This makes clouds the architecture of choice." 

Malik also reports a huge up-tick in desktop virtualization, which is another example of a dynamic workload. "For new trading floors, centralized compute resources are leveraged from the datacenter with thin clients installed on the desktop," he says, citing one firm testing desktops in London serviced from the eastern US. "Internet service providers (ISPs) are exploring this market to manage all of the necessary resources with a consumption based billing model for various user profiles," says Malik. "We can easily drive 20 to 30 hosted desktops for back-office users from a single node."

The List Group, a consultancy based in Italy, is currently building out its risk and trading solutions to exploit grid architectures or even multi-core systems on a single box. "While we can distribute risk calculations across the nodes, trading is particularly sensitive to latency, so we need to keep the processing on one backplane or physically close," says Enrico Melchioni, managing director of FMR Consulting at List Group. "Sharing data across the cloud and supporting a low-latency messaging layer add further challenges for clouds. However, having converted our software to highly parallel processing, we are clearly well-positioned to support whatever brokers may decide to offer."

Melchioni, however, does see an opportunity for clouds in proximity hosting at the exchange, with software driven resource requests that could scale with client demands for algorithmic trading, for example. "Brokers could maintain a minimal footprint and scale up as required," he says.

Endless Possibilities

In speaking with financial services, IT professionals, the cloud concept appears to have released a wealth of creativity. "Cloud is very descriptive," says Majedie's Hazlitt. "We live in a public world where 90 percent of information exchanges of value are across company borders, and because it is all browser-based, we can use our systems from anywhere on earth, from Majorca, for example, as easily as from London." His firm calls it "running on air."

For Hazlitt, not having internal IT actually lowers the barriers to entry for many new business areas. "Cloud computing will accelerate that, but more than that, it changes the model," he says. "The bar to value creation has just dropped. You can test out a concept on Amazon's Web services for $12.95."

Malik at Cisco is equally enthusiastic: "Every C-level executive should be thinking about clouds with a SaaS offering, which provides a rich user experience, reduces cost, time-to-market and increases asset utilization. Some financial institutions are even thinking about how to enable their grids into revenue generators by selling off surplus capacity to hedge funds in an infrastructure-as-a-service model."

"During the boom the pressure was always on delivering product," says Kelso at Deutsche. "Now in the crisis, we have to consider fundamental change. Budgets can be restrictive but the demands have not gone away."

Bob Giffords is an independent banking and technology analyst. He can be reached at bob.giffords@btinternet.com.

-->