The most useful machine in a telecommunications laboratory may be the one pretending to be something else. A switch, perhaps. A wireless base station. A network database that answers too slowly, or incorrectly, or with a message arriving several beats out of sequence. Catapult Communications built those machines. For nearly a quarter-century, it gave network engineers an artificial world in which things could go wrong before the real world was invited in.
This was an oddly cheerful business built around bad news. Catapult's DCT2000 and MGTS systems simulated pieces of a phone network and tested whether new equipment behaved properly. They checked features. They checked conformance to changing standards. They checked whether one vendor's box could talk to another vendor's box. Then they turned up the traffic and watched whether the system slowed gracefully or collapsed.
The thing they actually sold
A laboratory full of convincing impostors
Richard A. Karp, a Stanford-trained computer scientist, founded Catapult in 1985. Its first DCT system arrived that year and kept evolving as telecommunications lurched from one alphabet soup to the next: SS7, GSM, GPRS, CDMA, UMTS, IMS, WiMAX and eventually LTE. The labels changed. The basic anxiety did not. A new network component had to communicate with everything around it, including machines that had not been designed by the same people, in the same country or even in the same decade.
One box, four kinds of uncomfortable question
Catapult packaged proprietary software, programming tools and protocol libraries with modular hardware. DCT users could build tests in a graphical interface called CATTgen, write in Catapult's own programming language, or bring C and C++ routines. MGTS offered a graphical state-machine builder. Desktop systems held a few interface cards; the rack-mounted m5000 connected as many as 18 CompactPCI cards for heavy 3G load tests.
A configured sale typically ran from about $50,000 to more than $250,000. Customers licensed protocol modules, bought the interfaces their lab required, then added modules, hardware, maintenance, training and consulting as needs changed. This was less like buying a hammer than subscribing to an encyclopedia whose newest volume explained how the next mobile standard could misbehave.
The customer list was the credential
Everyone in the room already knew the acronyms
Catapult sold to the people building networks and the people operating them: Alcatel-Lucent, Ericsson, Motorola, NEC, Nokia Siemens Networks, Nortel, Fujitsu, NTT DoCoMo, AT&T Mobility, France Telecom and Vodafone appeared among its customers. The sales force was small and technical. A purchase could require months of evaluation and capital approval, but one installation could lead to another lab, another division or another geography.
The company's real expertise was not a particular chassis. It was accumulated exception-handling. Standards describe how protocols should behave; markets are full of regional variations, manufacturer quirks and legacy decisions. Catapult maintained modules for hundreds of protocols and variants. The larger the library became, the harder it was for a newcomer to reproduce the strange little facts embedded in it.
That independence mattered. Catapult did not need to protect a competing switch or base-station business. Its rivals included specialists such as Artiza and NetHawk, instrument makers including Agilent, JDS Uniphase, Spirent and Tektronix, and an awkward category found inside the customer's own building: internal test teams. Catapult's pitch was breadth, speed and neutrality. One configurable system could replace a small zoo of narrower tools.
What failed first
The network held. The customer base compressed.
The first dangerous failure was not technical. It was concentration. In fiscal 2008, Catapult's five largest customers produced 54 percent of revenue. Alcatel-Lucent alone supplied about 16 percent; NTT DoCoMo and Ericsson added 12 and 11 percent. When large telecom suppliers combined, duplicated development groups disappeared and test purchases slowed. Customers in Japan also had their own equipment offerings. Catapult's revenue slipped 4 percent to $37.9 million, and the company posted a $7.4 million operating loss.
Karp changed his mind when the order rate failed to improve in the first quarter of fiscal 2008. He chose to cut the spending rate while still funding LTE development. Catapult eliminated 25 positions - 11 percent of its workforce before the restructuring - and closed its Australian R&D operation. The move cost $2.2 million. It was not a retreat from engineering so much as a choice about which engineering deserved the next dollar.
LTE did. Orders began to appear in the final quarter of the year, including from equipment makers in every major Catapult sales region. The company had learned a cyclical lesson from an earlier downturn: preserve enough balance-sheet strength to invest when customers return, then broaden the business once the new cycle is moving. By March 2009, quarterly revenue had risen to $12.1 million from $9.9 million a year earlier.
The buyer saw the missing half
Wireline met wireless for $104.6 million
Ixia announced its offer in May 2009. It was strong in IP performance testing and service verification; Catapult was deep in wireless protocols. As fixed and mobile networks converged, the product map had an obvious empty space. Ixia offered $9.25 in cash per share. On June 23, about 95.5 percent of Catapult's outstanding shares had been tendered, and the merger closed. Ixia paid roughly $104.6 million for the shares, plus $2 million for eligible options.
The combination became IxCatapult. Ixia later described it as an in-depth testing line for 2G, 3G and LTE access and core components, linked with IxLoad for mobile-core testing. In 2017, Keysight acquired Ixia. The old Catapult products eventually moved into an end-of-life archive, but the underlying idea traveled much farther than the wordmark.
The lesson worth stealing
Build the rehearsal before the performance
Catapult's method is easy to borrow even when there is no base station in sight. First, model the surrounding system, not just the object being tested. Second, encode known standards and the awkward exceptions learned from customers. Third, inject wrong order, wrong timing and excessive volume. Finally, inspect the shape of the failure. A graceful slowdown is evidence. A silent cascade is a warning.
This works best where mistakes are expensive, environments are repeatable and customers share enough underlying standards for a library to compound in value. It works less well when failures cannot be reproduced, interfaces change faster than tests can be maintained, or every deployment is so bespoke that yesterday's exception teaches little about tomorrow's. Catapult also demonstrates a business limit: a superb test product does not protect its maker when a handful of buyers consolidate.
Still, the central proposition remains wonderfully plain. Catapult did not promise that complex systems would never fail. It offered a private appointment with the failure first. In a business full of futuristic acronyms, that may have been the most practical technology of all.