ASPEED Software (www.aspeed.com), a Microsoft partner, enables firms to rapidly upgrade their applications to fully exploit multi-core and grid, significantly shortening response time. By adapting Windows- and Linux-based single-thread desktop and server applications, explains ASPEED’s CEO Kurt Ziegler, customers can take full advantage of Intel-based multi-core platforms without major application surgery or retooling. We recently caught up with Kurt just as he finished meeting with Richard Dracott, general manager, high-performance computing group at Intel (www.intel.com).

WFS: Kurt, what do you mean when you say that you can ‘fully exploit’ multi-core and grid?

KZ: Multi-core provides the equivalent of multiple CPUs for an application’s use. The challenge is that almost all current desktop applications and most server applications are either entirely or partially single-thread logic. While different applications can use the cores simultaneously, specific portions of a given application can only use a single core at one time. This serialization is even true for applications that have been partitioned to run on a cluster or grid since the gridified application portion is still single thread.

WFS: Apps that run as a single thread or rely on libraries that run in single thread can only use one core at a time, limiting the response time to the speed of a single CPU. How can developers speed up the application without redesigning the entire application?

KZ: Two solutions. With our software, you can wrap binaries, such as DLLs or XLLs, or insert our APIs in the source code around one or more portions of the program that lend themselves to be run in parallel. By wrapping the binaries, the application logic is not changed and our ASPEED software provides the parallelization and data-distribution management. The API option enables the programmer to instrument the application while preserving the program logic without worrying about the distribution logistics.

Either way, we take care of all the parallelization and distribution logic and dynamically optimize the execution in real-time to ensure that cores are being optimally used.

WFS: Many firms use Excel for many facets of their activities. Given that Excel 2007 provides for multi-threading, is there anything additional that ASPEED provides for the Excel applications?

KZ: Quite a bit. Microsoft 2007 has parallelized the recalc engine and provides for parallel thread-safe applications to run on a Microsoft SharePoint Server. We enable Excel 2000, 2003 and 2007 apps using VBA and single-thread UDFs to fully exploit multi-core systems, servers, clusters and grid platforms.

WFS: To what effect?

KZ: Many of our customers reduced long running Excel applications to minutes, while others significantly reduced transactional times for pricing and risk assessments using proprietary UDF library functions. We also complement Excel 2007 by facilitating the co-existence of the new Microsoft HPC functionality with legacy VBA and UDF libraries, avoiding the need to re-write their libraries with thread-safe code.

WFS: How does ASPEED’s approach to functional distribution differ from what the grid platforms do?

KZ: The best analogy is to liken grid platforms to air traffic controllers and ASPEED to airplane pilots. Grid infrastructures schedule and provision capacity minimizing queue times by workload leveling. To reduce a specific application’s run time, it is typically carved into discrete steps passing dependent data among the various steps. ASPEED complements and enables grid by focusing on the performance of individual applications. Each application dynamically distributes its workload among the resources provisioned by the grid, allowing the application logic to run multiple processes on each worker/engine/service provided, reducing elapsed time and improving the effective utilization. Applying ASPEED to a gridified application or running an ASPEED gridified application provides additional response time reductions of 15 to 300% with associated reduction in utilization, freeing up the resource for other work.

WFS: One more question. Can ASPEED help firms that have major investments in Unix-based multi-processing applications scale with the business volumes using a multi-core Linux-based cluster or grid?

KZ: Yes, several of our clients moved their applications off SMP-based UNIX in order to scale on Intel-based Linux-based clusters and grids.

WFS: Thanks, Kurt. Let’s turn to Richard. Why are firms demanding an exponential growth for computing capacity and how does Intel help?

RD: Computers play an increasingly important role in the front office, allowing Wall Street firms to create more products, make more trades through faster trading, faster modeling and to compute risk faster, what we call faster risk. These firms spend about $1 billion a year on research and development learning how to improve computing risk, allowing them to set accurate prices for options and other derivatives.

Over the past five years, there’s been rapid advancement bringing better clarity on how to monetize compute cycles, which has driven the need for faster trading and faster risk.

For example, a few years ago, big iron SMP boxes started running out of gas and Wall Street sought new solutions delivering more compute cycles, leading firms to take a serious look at distributed solutions like those at ASPEED, an Intel partner, because they can exploit the ‘embarrassingly parallel’ characteristics of many Wall Street applications while managing the dependent data aspects, quickly taking them from the SMP world, adapting them to Intel-based distributed clusters.

ASPEED also offers the benefit of being complementary with grid solutions where firms get the benefits of Intel-based distributed processing today, even before they pick their grid strategy.

WFS: How so?

RD: ASPEED allows legacy, single-threaded apps to run in multiple instances, dynamically balancing the right number of instances for the number of compute resources available. Paired with Intel’s quad-core Xeon processor, you get the most compute resources available per CPU today. Servers based on the new, low power, quad-core processors are designed for dense Internet datacenters with blade servers where the scale and density of servers are highly sensitive to power, real estate and cooling costs.

WFS: It’s not just performance though, is it? What’s Intel doing to deliver energy efficient solutions?

RD: Intel has energy-efficient, 50-watt server processors that represent a 35- to nearly 60-percent decrease in power from our existing 80- and 120-watt quad-core server products. These new processors represent a nearly ten-fold improvement in power consumption per core, compared to 64-bit Intel® Xeon® Processors at 110W per core just 18 months ago.

As firms focus on reducing electricity bills and cooling costs associated with computing, these new processors, while requiring just 12.5 watts of power for each of the four cores or processing engines, deliver similar performance yet set a new standard in energy efficiency.

WFS: How does Intel view the convergence of virtualization, grid computing and high-performance computing?

RD: Financial firms have dedicated HPC resources to handle the compute-intensive applications such as mathematical algorithms and modeling. They also look to virtualization to share resources in an SOA/grid model. In both cases, Wall Street demands more compute cycles in the same power per square foot envelope and Intel’s quad-core Xeon CPUs are out in front on that.

WFS: How can Intel help with performance around Microsoft?

RD: Microsoft’s financial Longhorn/Viridian/Vanderpool with Virtual Server running on a new Intel Xeon Quad core 4-way will enable users more flexibility when using Excel to analyze large enterprise-customer revenue and discounting trends. Those Excel books take hours to process on normal laptops and desktops but process nearly instantly when enough large hardware can be provisioned.

With new technologies in the virtualization space, the solution will enable more flexibility to step up power as needed for key financial analysts working round the clock.

WFS: Speaking of clocks, we’re out of time. Thank you, both.