TECHNOLOGY TRANSFER
ERCIM News No.37 - April 1999

Virtutech Commercializes SimICS

by Peter Magnusson


The SimICS simulation-based development tool was commercialized in 1998, with rights of development and exploitation being licensed to Virtutech AB, the first formal spin-off company in SICS’ history.

Software and hardware designers of high-end computer systems that play a key role in emerging complex information systems, ranging from database engines to telecom/datacom transaction-oriented systems, lack adequate debugging tools both as far as functionality and performance issues are concerned. Such tools would drastically reduce time-to-market thus improving competitiveness in a fundamental way.

Developed in the computer architecture group at SICS over a period of several years, SimICS is a “system level instruction set simulator”. It is a specialized development tool for computer architects and operating system developers that models a target computer at the level of individual instructions. It offers unique benefits in high-end computer system design work beyond traditional tools and methods. For example, SimICS can run a realistic commercial workload in a fully virtualized system, allowing perfectly repeatable measurements that can track bugs that violate specifications regarding function as well as performance.

Virtutech specializes in high-end full-system simulation technology. In the brief period since opening its doors in the summer of 1998, Virtutech has signed a series of key strategic customer agreements, including with Sun Microsystems, Ericsson, and Hewlett-Packard.

“The simulator group that founded Virtutech is one of the world’s foremost research groups in system level simulation”, according to David W. Yen, vice president and general manager of enterprise server products for Sun Microsystems, Inc. “Sun has worked with them for many years, and it is good to see they have discovered the commercial potential of their work and are beginning to capitalize on its worth.”

SimICS is now marketed under the label Virtutech SimICS, and is the first product sold by Virtutech, in addition to specialized consulting services. Virtutech expects to have a profitable first year of operation, and 1999 is already “sold out” with customer projects.

“The tools and technology that we provide customers are unique in the commercial space,” says Virtutech CEO Peter Magnusson. “Historically, this class of technology has been restricted to in-house proprietary efforts, deemed too sensitive and strategic to be outsourced. We have pioneered the concept of a specialized, independent company in the field. The technical and economical benefits of our specialization are significant. In todays competitive world, no stone remains unturned in improving quality and time to market. Our first customers are exemplary leading edge companies, early adopters of new technologies and processes.”

It began with Architecture and a Misunderstanding

The work leading up to SimICS originated in computer architecture and operating systems research at SICS in the late 1980s and early 1990s. In projects led by Seif Haridi, Erik Hagersten, and Andrzej Ciepilewski, it was observed that the existing simulation techniques failed to capture significant characteristics of ‘real’ commercial systems. A tradition of focusing on scientific workloads had led to a general feeling in the field of computer architecture that available benchmarks and techniques had systematic methodological weaknesses.

A computer architect uses simulation to provide input into his or her architecture models. Much of computer architecture involves careful optimization of common cases, perhaps to the detriment of less common events. The relative frequencies of events in real uses of computers are thus important both to guide design choices, and to predict their relative impact. Because of limitations in existing simulation techniques in the late 1980s, computer architects in academia were restricted to models driven by (a) scientific, rather than commercial, programs, and (b) only so-called ‘user-level’ events, as opposed to operating system involvement. By contrast, there were indications that typical commercial workloads, such as large databases, behaved rather differently. In later years, this was confirmed by detailed studies performed by IBM, among others.

The first directions in the work leading to SimICS was based on a misunderstanding. An existing, ground breaking simulation environment developed by Robert Bedichek at the University of Washington was extended to support a multiprocessor system and to mimic real devices of a prototype research architecture, the SICS DDM. The first programmer on the project developed simulator device models sufficiently accurate to boot and run completely unmodified operating systems, including commercial on-board firmware. This strategy was chosen on the - incorrect - assumption that it was common practice. The strategy proved cumbersome, and eventually it became evident that SICS was the first open research facility to seriously undertake such a programming task. The early programmers on the project were confident that the task was not only feasible, but would take around six months or so. The obvious benefits of running a real, commercial workload, led a series of research managers at SICS to support continued work in that direction.

Some six calendar years, twenty man years, and several hundred thousand lines of code later, in 1997, the simulation group in the Computer and Network Architectures (CNA) group at SICS finally succeeded in the original goal: booting a commercial operating system (Solaris 2.6) on a simulated Sun Microsystems server (sun4m architecture). This was the first known occasion of an academic group running an unmodified commercial operating system in a fully simulated environment.

In the meantime, the architecture research at SICS achieved great success, and significantly influenced commercial designs. SimICS has followed in the wake of SICS computer architecture results, and has now also taken the step into the commercial field.

The simulation group at SICS eventually grew to five people, all of whom became founding employees of Virtutech: Magnus Christensson, Fredrik Larsson, Peter Magnusson, Andreas Moestedt, and Bengt Werner. The board of directors of Virtutech include Bo Hedfors, an earlier chairman of the SICS Foundation, and Stig Larsson, the current chairman, as well as Per Stenström, professor of computer architecture at the Chalmers University of Technology.

Other than those already mentioned, significant contributions to the success of simulation work at SICS over the years were made by Torbjörn “tege” Granlund, Anders Landin, and David Samuelsson.

Virtutech is located in downtown Stockholm. For more information, see http://www.virtutech.se and http://www.simics.com

Please contact:

Peter Magnusson - Virtutech AB
Tel +46 8 690 0720
E-mail: info@virtutech.se


return to the ERCIM News 37 contents page