The Aegis Weapon System represented a triumph of computer programming when it became operational in the early 1980s. It had five million lines of computer code, more than the Space Shuttle or even the super-secret B-2 bomber. The program also was reliable—as it had to be to defeat supersonic sea-skimming missiles.
Recent upgrades have increased the system's computing power. The current Baseline 6 upgrade alone adds 15 million lines of code while maintaining the system's reliability, which must grow even more to meet tomorrow's threats. Its aging mainframe computers, built to military specifications (MilSpecs), are saturated and remain frozen in time. Meanwhile, commercial off-the-shelf (COTS) equipment continues to advance according to Moore's Law, which states that the number of integrated circuits that can fit on a given-sized computer chip will double roughly every 18 months. The Aegis Program Office already is applying this burgeoning commercial processing power to perform new functions like theater ballistic missile defense.
"Adjunct processing" is helping to get commercial equipment into the system as soon as possible. This interim approach grafts commercial processors onto the system to perform specific tasks. Of the 15 million new lines of code in the Baseline 6 upgrade, 10 million will run on adjunct processors. But while it can add a lot of capacity, adjunct processing has serious limitations. Commercial hardware and software can never realize their full potential while bound to a core of old MilSpec mainframes. More important, commercial equipment cannot assume the core weapon-system functions as long as the computer programs that perform those functions run only on machines hardwired to one another by a maze of cables. This rigid layout, or architecture, is as much of an obstacle as the old mainframes themselves.
The first step toward a true commercial system was to install all the millions of lines of code on a new computer architecture designed to accommodate the latest commercial equipment.
In June 1991, the Aegis program office teamed up with the Defense Advanced Projects Agency (DARPA), the Naval Surface Warfare Center's Dahlgren Division, and Johns Hopkins University's Applied Physics Laboratory (APL) to found the High-Performance Distributed Computing (HiPer-D) program. DARPA funded the first three years, and the program office the next three years. Lockheed Martin Corporation, long the Aegis prime contractor, established close ties with HiPer-D to translate the lessons learned into improved combat performance.
In developing a completely new computer architecture, the team followed the proved Aegis engineering principle of "build a little, test a little, learn a lot." It set up a permanent working laboratory at Dahlgren, Virginia, where engineers could test various commercial technology in different layouts; the first new architecture flopped, but it planted the seeds of later success. DARPA wanted to see how well a new supercomputer, Intel Corporation's Paragon, could perform combat functions. HiPer-D connected this mainframe to workstations with a conventional Ethernet communications backbone and installed redesigned Aegis software. In March 1994, an initial (II) test proved conclusively that this mainframe-based commercial architecture could not meet Aegis combat requirements.
The architecture could not communicate fast enough to take advantage of the supercomputer's great speed: It could handle only enough data to keep track of about 50 airborne objects at a time. Beyond 50 simultaneous tracks, it tended to go into cardiac arrest. In comparison, the old MilSpec-based architecture now in the fleet can track more than 700 prospective targets simultaneously. The whole process was a valuable lesson that confirmed the Aegis program's longheld belief in total system engineering by demonstrating, once again, that the key to meeting combat requirements is not the speed of any one piece of equipment, but the speed of the entire system. It convinced DARPA's researchers to shift their focus to what they call "embeddable systems," which deemphasized pure computing power in favor of communications hardware and software that could meld many commercial processors into a capable network. As the "D" in HiPer-D foretold, the future lay in distributed computing.
While DARPA pursued technology that would help support fully distributed computing, HiPer-D, which was now being funded by the Aegis Program, concentrated on developing a partially distributed architecture that would permit the weapon system's computer programs to run effectively on commercial equipment. This new system incorporated commercial network technology and adopted some of the features found in commercial distributed architectures. Instead of a few mainframes, it had many commercial processors tied together by local area networks (LANs), each of which used a commercial fiber-optic data bus to move a large volume of data flexibly. High-speed commercial routers transferred messages from one data bus to another, permitting all of the networks to work together as an integrated weapon system.
In May 1995, HiPer-D gave this partially distributed architecture its first major test, designated TI. The new system ran the existing Aegis computer program extremely well, maintaining and distributing just as many radar tracks in real time as the existing MilSpec system. It also demonstrated a fault-tolerance capability that ensured reliability. If a fault developed, the system automatically brought in backup capabilities with no perceptible interruption in service. The success of TI showed that the time was ripe to move toward production of a commercially based Aegis weapon system.
Representatives of the program office, Dahlgren, APL, and Lockheed Martin met in July of 1995 to kick off an engineering development program based on the new commercial architecture. The program centered on Engineering Development Model (EDM)-5, the fifth such prototype in Aegis history. Now installed at Lockheed Martin's Computer Programming Test Site in Moorestown, New Jersey, EDM-5 has more than six times the processing power of any Aegis weapon system now in existence.
An early demonstration on 18 December 1996 showed that EDM-5 can run all of the current Aegis computer programs, including the programs for command and decision, weapon control, fire control, and the SPY-I radar.
All of those programs had been translated from the old CMS-2 programming language to Ada, a language common to many advanced weapon systems in the United States and abroad. With the translated programs, EDM-5 succeeded in tracking a great many targets, identifying them, determining how to engage them, controlling the engagement, and determining whether to re-engage them—all while supporting numerous operator consoles and displays and constantly monitoring the entire system.
As a working prototype, EDM-5 is giving combat system engineers ample opportunity to apply total system engineering, ensuring that despite the transition to commercial systems, the Aegis weapon system will function as effectively and reliably as ever. The new architecture will eventually become part of the Baseline 7 weapon system, which is scheduled to go to sea in the DDG-91, the 41st ship of the Arleigh Burke (DDG-5 1) guided-missile destroyer class. The Navy plans to award a contract to build DDG-91 this fiscal year, and the ship is scheduled join the fleet in fiscal year 2003, a mere six years from now.
While work progresses on EDM-5, the HiPer-D program continues to expand the bounds of Aegis computing, which is moving rapidly toward a fully distributed architecture. Because it must run a translation of computer programs developed for the existing Aegis weapon system, EDM-5 still reflects the basic design of the old MilSpec system. Each of its local networks, for example, plays much the same role within the system as the equipment it replaces.
In comparison, the computer programs that run on the next-generation system will not only be written in Ada, but also will be restructured to derive the maximum benefit from a distributed architecture. Fault tolerance, always an Aegis hallmark, will improve. Whereas today's hard-wired Aegis weapon system has but two alternative paths for performing each critical function, a fully distributed system will provide many alternative paths and will be able to switch paths virtually instantaneously. Adjustments will take place without the slightest interruption in service.
A distributed architecture also will help break down the wall that separates combat functions and other shipboard data processing. A large number of processors distributed throughout the ship will be equally capable of performing combat functions such as maintaining tracks on incoming missiles and non-combat functions such as sorting through personnel records. In theory, this could make the number of alternate paths for fault tolerance virtually infinite. In combat, of course, the system will perform critical functions only on a large but finite set of paths whose effectiveness and reliability have been tested thoroughly beforehand. The distributed architecture's automatic resource-management feature will reconfigure only the computing resources assigned to support combat functions when there is no imminent threat. That should guarantee enough time for rigorous testing of every new backup path.
A good measure of progress is the growing scalability of each successive architecture. A fully distributed architecture allows designers to scale up, adding processing power or introducing new technology without significant disruption. Commercial standards help ensure that the added hardware and software is compatible with what came before, even if the additions represent a new generation of technology. Advanced data-distribution technology, such as fiber-optic data buses and high-speed message routers, are themselves easily scalable to handle the additional traffic.
The scalability of successive HiPer-D architectures is illustrated by their simultaneous target-track capability. In the 1994 I1 demonstration, the rigid commercial mainframe system threatened to crash beyond a meager 50 tracks. In contrast, the partially distributed system in the March 1995 Ti demonstration could maintain more than 700 tracks, the same as the best MilSpec system; EDM-5, which evolved from T1, can handle a somewhat larger number. EDM-5's architecture, however, which is only partially distributed, still limits its scalability. In the November 1995 T2 test, HiPer-D demonstrated a new architecture that approaches full distribution. During T2, this system easily handled 1,400 tracks, double the current capacity of the Aegis weapon system.
HiPer-D then began to incorporate the Myrinet advanced communications hardware that had emerged as one of the fruits of DARPA's embeddable systems work. DARPA's Myrinet, together with other commercial network technology introduced by HiPer-D, enabled the system to achieve yet another increase in capability. In a flawless follow-on test (T2A), on 19 December 1996, the system maintained 3,000 simultaneous tracks without a hitch—and gave strong indication that it would be able to handle many more.
T2 and T2A also furthered HiPer-D's relationship with DARPA, helping the agency's technologists focus the agency's new "Quorum" effort. Weapon systems developers strive for the speed and reliability required for combat by custom-designing every interface between major elements of the system, even though this restricts the flow of data and limits power and flexibility. In contrast, those who develop commercial distributed systems achieve great power and flexibility using an open client-server environment, in which any processor can obtain whatever data it needs from anywhere else in the system. This openness, however, has always meant some loss of speed or reliability, which a combat system cannot tolerate.
Quorum is developing middle-ware, to coin a term, that will free the high-level applications that perform combat functions from having to deal directly with the low-level operating systems that control each individual processor. This will allow the high-level applications to run faster and more reliably, permitting an open, client-server architecture to guarantee the same level of performance as any custom-designed system.
As DARPA develops this Quorum technology, HiPer-D will continue to incorporate it into a fully distributed architecture continuously upgraded with the best hardware and software available on the commercial market. Progress to date indicates that fully distributed computing will be ready for engineering development within the next few years. Just as today's EDM-5 prototype has more than six times the power of the Aegis weapon systems now entering the fleet, so the fully distributed system will be more than six times as powerful as EDM-5.
The open computer architecture now being pioneered by the Navy, HiPer-D, DARPA, and the other organizations that make up the Aegis community, ensures that the service will be able to take advantage of ever-more-powerful commercial hardware and software as it becomes available.
Captain Meyer is the Aegis Technical Director. Captain Geary is the Branch Head, Aegis Combat System Engineering Branch.