NT may rule the desktop, but UNIX remains dominant for servers.
Is NT reliable and scalable enough for risk management?

By Andrew Webb

One could be forgiven for thinking that since Windows NT captured the financial market’s desktop in barely 36 months, it would make pretty short work of grabbing the mission-critical server side as well. Yet strangely enough, we aren’t being deafened by the sound of UNIX CDs being flung into the trash can. NT may have almost completely taken over the client side of risk management, but when it comes to the heavy-iron end of the business, the banking industry appears curiously reticent to trust NT with its risk management database and, in many cases, its risk management application servers. Although Microsoft is throwing some serious resources at this one, UNIX still seems to be hanging tough.

The principal problem with NT is its reliability. Financial technologists can accept the profanity and missed opportunities that follow a trader’s desktop crashing here and there. But if the operating system running an enterprise-wide risk database falls over, the entire organization could be left drifting rudderless towards oblivion in a volatile market.

The other important issue is scalability. The standard release of NT Server can support only four processors. Custom non-Microsoft kernels, which support a larger number of processors, are available from Tandem and others and come at a price. “In areas such as risk management, where processing requirements are already substantial and continue to rise, I just don’t see Microsoft making the necessary leap,” says Virginia Gambale, managing director and CIO at BT Alex Brown. “Though it recovered magnificently when it misjudged the browser market, I don’t think one can apply the same paradigm here.”

These shortcomings don’t appear to be simply a triumph of perception over reality. “The NT community is clearly tooling up to demonstrate true enterprise-wide scalability and deployment. Until this message is driven home, it is natural that some people will remain circumspect,” says Richard Walker, vice president at Infinity. “Individual benchmarks or tests can be focussed on just one aspect and fail to take into account how something will be deployed across the entire enterprise.” As a result, Walker believes, the risk management marketplace will only sit up and take notice when NT is seen to work in mission-critical applications in a fault-tolerant, scalable environment.

“In areas such as risk management, where processing requirements are already substantial and continue to rise, I just don’t see microsoft making the necessary leap.” —VIRGINIA GAMBALE BT ALEX BROWN

Others, such as Roger D. Lang, risk solutions marketing manager at Compaq, disagree. “I think customers are starting to gain confidence in NT for risk management as more applications come on stream,” he says. More specifically, Lang points to companies such as Barra as indicative of a trend among younger risk management vendors toward using NT throughout the system, rather than only on the front end. Lang contrasts this with the lack of newcomers building new risk management applications from scratch to run on UNIX. He credits much of UNIX’s remaining presence in risk management to established risk vendors who have a historical commitment to the operating system and who are understandably unenthusiastic about the task of porting the server side of their applications to NT.

Wobbly platform

How reliable is NT? For some, NT’s stability shortcomings render it completely unsuitable for the time being as a server for mission-critical applications such as risk management databases. “I think the terms ‘high availability’ and ‘NT’ are mutually exclusive at the moment,” says Alistair Mortimer, head of systems strategy for Abbey National Treasury in London, one of the world’s largest nonsovereign issuers of debt. “NT leaks memory, has weak handling of third-party device drivers and still suffers from dynamic-linked-library conflicts, with version control and support in terms of layered products also quite poor. When you hear Microsoft and Compaq talking about ‘graceful degradation,’ this seems to mean the machine showing a few warning lights before it crashes rather than falling over instantaneously, and it doesn’t inspire confidence. We have grown to accept that our desktop PCs will crash perhaps as often as once a day, and currently server availabilities are not that much better, which hardly inclines one to trust NT right now with something as critical as a risk database.”

Others believe NT’s chief structural weakness is the peer-to-peer relationship between server and workstations. “That makes it far easier to bring down an NT system accidentally,” says John Best, program manager in IT development at Westdeutsche Landesbank in London. “As a result, a lot of traditional mission-critical systems such as risk management have UNIX back ends with NT at the front—which provides additional security and stability because it’s not a peer-to-peer relationship.”

To be fair to NT, the playing field isn’t exactly level when it comes to hardware. When buying a UNIX server for a risk management system, “reassuringly expensive” seems to be a common philosophy. “The reliability of the boxes on which the operating system runs is still tilted in favor of UNIX—people are prepared to spend a bit more on UNIX,” says Best. “Not everyone runs NT on the best hardware.” Best also highlights the common temptation of using an under-specified NT box, which can compromise the stability of the environment, since only one or two underperforming servers can have an enormous impact on the overall throughput and resilience of a system.”

“You’ll put eight or 16 gigabytes of memory and maybe eight processors into a UNIX box without giving it a second thought,” says Vladan Jovanovic, CEO of trading and risk management at Communicating Ltd. in London. “Compare that with the general approach to NT machines, where the emphasis is on finding the cheapest box and shoving it under someone’s desk where it accidentally gets kicked five times an hour.” While NT servers intended for risk management may not be treated in such a cavalier fashion, this sort of treatment is commonplace where NT is used for less-critical server applications such as file and print serving. At least part of NT’s reputation for reliability problems can be attributed to this.

UNIX has gained some of its reputation for reliability simply because it has been around longer and has had time to bed down. Twelve years ago, many of the criticisms being leveled at NT today were being leveled at UNIX. “There’s been plenty of time for large, bullet-proof implementations of mission-critical applications to evolve on UNIX,” says David Penny, chief technology officer at Algorithmics. “Although Microsoft has a lot of top-flight personnel working on it, there’s an inevitable shaking-out period.”

“The smaller address space and limited parallel-processing capability of nt still permits one to serve the risk management needs of the vast majority of mid-size corporate and banking portfolio managers.” —MICHAEL KOWALSKI ASKARI

Another factor has been the complexity and ‘unfriendliness’ of UNIX when compared with NT. Perversely, NT’s graphical user interface has made it all too easily accessible to the wannabe technician. Snafus inevitably result from amateur attempts to ‘configure’ NT, and the blame often falls on the operating system. The same people, however, would be far too intimidated to attempt to configure something as incomprehensible as an expensive UNIX box—which they wouldn’t be allowed near anyway.

With ever-increasing pressure on IT budgets, however, the issue of hardware costs could prove NT’s ace in the hole when it comes to large applications such as risk management. With Microsoft promising great things on the clustering front, the possibility of distributing fault-tolerant applications across a farm of commodity Intel boxes running NT will be quite cost-attractive. “If you wanted to do something like run Monte Carlo simulations in real time, this would make good economic sense,” says Vincent Aubrun, director of technology at Summit Systems. “There’s no mass market for high-end UNIX boxes, so you don’t have the potential for prices falling to a level that could compete on a cost basis with such a setup.”

Some vendors highlight extremely specific areas of NT that they feel make it currently unsuitable for risk management applications. “There is something definitely awry with NT’s TCP/IP stack and networking,” says Andy Abrams, executive vice president of software development at trading and risk management vendor Kestrel in New York. “You can take the same piece of hardware and put NT on it and then compare it with Solaris [Sun’s UNIX flavor]. The difference in network performance is staggering. In a mission-critical environment, such as risk management, the throughput from net to card to CPU needs to be good, and on NT it simply isn’t.”

How scalable is NT?

For the largest banks, NT’s lack of scalability when compared with UNIX pretty much rules it out as a serious risk management operating system. The huge portfolios of some of the largest players are simply beyond its scope. However, a recent benchmarking test of Infinity’s Panorama (the first end-to-end NT risk management application) at Dell Computer Corp. Application Solution Center in Limerick, Ireland, showed that the potential is at least there. A 261,000-trade portfolio was distributed among 11 workstations (10 dual-processor and one mono-processor Pentium II Xeon) and valued in 9 minutes. A Monte Carlo value-at-risk using 2,000 scenarios took one hour, one minute and 18 seconds to produce.

Of course, apart from the largest players, the need for scalability and the ability to address large quantities of memory are less imperative. “If risk management is considered primarily in terms of number-crunching VAR for the largest banks’ global portfolios, then the four-GB address space of NT is not adequate, assuming one insists on holding the entire portfolio in memory,” says Michael Kowalski, software development manager at Askari. “However, the ability to accommodate 200,000 positions safely within the smaller address space and limited parallel-processing capability of NT still permits one to serve the risk management needs of the vast majority of mid-size corporate and banking portfolio managers.” While these managers don’t have globally deployed database infrastructures and don’t have the resident expertise to administer UNIX in all its complexity, they still need flexible risk management systems.

“At present, nt can’t catch unix when it comes to scalability. it would be questionable to use it for something that requires intense, real-time, multi-threaded calculation routines.” —TOM NUNN BANKBOSTON

Nevertheless, NT still has a way to go to catch UNIX in the scalability stakes. “At present, NT can’t catch UNIX when it comes to scalability,” says Tom Nunn, head of technology for global capital markets and treasury at BankBoston. “For example, it would be questionable to use it for something like our TRMS risk management system, which requires intense, real-time, multi-threaded calculation routines.” Although BankBoston has already largely moved to NT on the desktop, and is closely following its progress on the server side, Nunn doesn’t see the bank cashing in its UNIX chips on its risk management or other critical trading-room servers in the immediate future. He does, however, see NT beginning to close the gap on UNIX in certain areas, such as tools for remote management.

The critical structural scalability problem for many technologists is NT’s lack of true multinode clustering. This is technology that has been commonplace in the UNIX world for some time and combines multiple servers into one logical unit that is transparent to the user and has true load-balancing between machines. “Microsoft has been promising this for a while, but it hasn’t appeared yet,” says Communicating Ltd.’s Jovanovic. “Microsoft’s existing Wolfpack technology isn’t the real thing, since it only works between two servers and shared storage.” In Jovanovic’s view, one of the major obstacles to Microsoft’s development of true multinode clustering is one of the features that made it a winner on the desktop—namely, its GUI. “That’s a major overhead in programming terms, and if they dropped it, they could probably achieve true clustering relatively quickly.”

The Linux Question As a further irritant for NT, there’s a new kid on the block. While everyone is probably sick of hearing Linux being touted as the answer to everything up to and including the bubonic plague, the fact remains that it seems to be gaining credence extremely quickly. Whatever the political motivation behind it, a direct investment by Intel and the announcement of Linux ports of existing database products from Oracle, Sybase, Informix and Computer Associates is not to be sneezed at. Apart from the usual UNIX virtues (of which Linux is another flavor), Linux enjoys the additional cost advantage of being able to run on commodity Intel boxes just like NT. “A year ago, I would have thought it unlikely, but now I think there is the distinct possibility that Linux could give NT a hard time in the risk management server arena,” says Larry Tabb, group director of the securities and investments practice at the Tower Group. “It has the inherent benefits of Unix behind it, and if it starts getting support and there is a stabilized version that performs well and can be maintained, it has a fighting chance as a high-end server and risk-management-type OS.” Tabb also notes the interest and potential support from other large vendors such as Hewlett Packard. If Windows 2000 isn’t as reliable as promised, Linux could be looking good. Despite the attempts by the likes of Red Hat Software to give Linux a respectable corporate image, others remain less convinced. “Yes, it’s a good OS, and yes, it’s free, but I don’t see it having a huge impact in risk management,” says Communicating Ltd.’s Jovanovic. “The stumbling block is that it is decentralized. While the kernel is maintained by Linus Torvalds, the way everybody else contributes to peripheral components makes keeping up with updates a horrendous task. How would you maintain it?” Compaq’s Roger D. Lang takes a similar view. “I think Linux will appeal to the quants who will relish playing esoteric games with its user-configurable kernel,” he says. “But I still see NT ahead of it for use by the vast majority of risk managers.” —A.W.

Microsoft, however, has been busy enhancing other areas of NT that will help boost its credibility as a heavy-iron operating system for risk management. Historically, one of its shortcomings has been the necessity of shutting down a server in order to change or add a device or driver. This has been a particularly important consideration for a server running a risk management database, where, for example, hard disks fill up rapidly and shutting down the machine to install further storage is not an option. Microsoft has made enormous strides here in the last year or two, so that it is now possible to perform about 50 different actions on-line (including disk changes and mirroring) that previously required a reboot.

The billion-dollar question is whether niggles about NT’s availability and scalability will disappear in its next incarnation as Windows 2000. In some respects, Microsoft’s scrapping of the NT 5 name and its replacement with three distinctly branded server variants of the newly badged Windows 2000 is possible grounds for optimism among those looking to use the new operating system for larger risk management projects.

Furthermore, in the past few months Microsoft has upped the specification of the proposed top-of-the-range server variant, called Windows 2000 Datacenter Server. When first announced in the fall of last year, it was slated to offer 16-way symmetric multiprocessing (SMP) and the ability to address up to 64 GB of physical memory. The latest announcement on the Windows 2000 site, however, promises 32-way SMP—still not as big as the largest UNIX deployments, but big enough for most purposes.

Thus far, Microsoft has failed to deliver efficient multiprocessing. “We currently aren’t getting the full scalable benefits in the NT environment,” says West LB’s Best. “Put four processors in the box and under NT you only get three or so processors’ worth of performance.” Best feels that once the marketplace starts to see truly scalable eight-way processing working well in a clustered NT (Windows 2000) environment, it will begin to take it seriously for the server side of risk management.

Microsoft claims that its new Datacenter Server release (and its smaller cousin, Windows 2000 Advanced Server) will integrate “high-availability clustering, network and component load-balancing to provide excellent system and application availability.” Leaving the hyperbole aside, if Microsoft can actually deliver on its numbers, the whole ballpark could start to tilt in NT’s favor. The current NT Server memory-addressing limit (using a standard kernel) is four GB, so the upgrade to 64 GB is a significant boost, particularly in view of the way prices for memory chips have plummeted over the past four years. Once an operating system is actually capable of addressing large blocks of memory, it becomes feasible on a cost basis to buy the memory so that complete portfolios can be loaded into memory for valuation and risk analysis. This dramatically speeds up valuation times, since chunks of data no longer have to be swapped to and from hard drives.

The problem is that Windows 2000 is well behind schedule, and there is a big question mark over exactly when Datacenter Server will actually ship. The more cynical have even suggested that the announcement of the larger Windows 2000 flavors doesn’t herald an imminent release at all, but is just another example of Microsoft’s FUD (fear, uncertainty and doubt) marketing ploy designed to postpone major UNIX-based projects.

Trouble ahead

Despite this, it’s not all sweetness and light in the UNIX camp. While UNIX may currently be ahead on availability and scalability, it can’t compete in other areas. Many of the newer risk management vendors are working with NT rather than UNIX because of the cost and availability of development tools. Making an inexpensive development environment available has been part of Microsoft’s well-proven strategy of courting the developer community.

The big question is whether pressure from the younger risk vendors will do anything to break the deadlock between the larger institutions and the established vendors. The current environment is a stalemate: The larger vendors say clients aren’t asking for NT back ends for their risk management applications in great numbers. And clients respond that risk management applications simply aren’t available.

If this logjam breaks, then the usual application-driven rules will apply—an operating system with no applications fades away. Mortimer at Abbey National corroborates this. “If two years up the line our risk management vendor suddenly dropped Solaris support, then we’d need to move the OS,” he says. “In the current year 2000 environment, people are buying, not building, in-house—so this carries additional weight.” Nunn at BankBoston, while acknowledging the virtues of UNIX, agrees. “Although we primarily run on UNIX servers, we have some running NT because certain vendors simply aren’t offering their applications in UNIX,” he says.

Algorithmic’s Penny points to another opportunity for Microsoft to push its way into the heavy end of risk management. “Both CORBA [Common Object Request Broker Architecture—the UNIX protocol for distributed computing] and DCOM [Distributed Component Object Model—Microsoft’s equivalent] support distributed computing and provide the same functionality,” he says. “Given the server issue, however, it is critical that Microsoft supports DCOM on the UNIX boxes.” Penny contends that if Microsoft can do this and maintain the quality of that support, it will enable DCOM to gain a solid hold on the market and, in the longer term, significantly enhance Microsoft’s attempts to get NT onto risk management servers.

Hardware costs aside, NT also has UNIX beat when it comes to maintenance. “UNIX may be mature and incredibly flexible as well as deeply scalable, but it’s also extremely expensive to maintain and incredibly configuration-sensitive,” says Abbey National’s Mortimer.

Nevertheless, the flip side of this is that UNIX’s long pedigree in mission-critical database environments has meant that over the years the environment has developed a large residual skill base. As a result, it is generally easier to recruit skilled personnel (at a price) to maintain an Oracle or Sybase risk database running on UNIX than an SQL database running on NT. Although Microsoft has put a lot of resources into closing this skill gap with training programs, it still appears to have a way to go.

Another big Microsoft advantage is the fragmentation of the UNIX camp. Three years ago, there were about a dozen flavors of UNIX floating around Wall Street. That aggravation pushed many IT departments to welcome the single-vendor NT Workstation with open arms. In the intervening period, however, this fragmentation—which was never so prevalent on the risk management server side anyway—has abated still further, so UNIX on Wall Street is now largely centered on the Solaris and HP-UX flavors, both of which are backed by serious support organizations.

On the face of it, NT appears to have a lot of ground to make up. For smaller operations, however, it still has a great deal to offer, especially since effective risk management requires the easy and flexible integration of data from diverse sources. The interoperability of DCOM components with Excel and Access is not merely a GUI user-convenience issue. Flexible presentation is vital in providing insights that go beyond mere VAR, net present value and the Greeks, so it can be argued that flexible data integration is potentially more important than raw speed. “If you want to build a programmatic B-52, the UNIX environment is the right choice,” says Askari’s Kowalski. “If risk management has a large component of guerilla warfare about it, then NT begins to look more attractive.”

Was this information valuable? Subscribe to Derivatives Strategy by clicking here!