Andy Bechtolsheim and Rex Pugh were not Asynchronous Transfer Mode (ATM) believers during 1997, unlike Fore Systems’ CEO Eric Cooper. In The Upstart Startup: How Cerent Transformed Cisco (2014), Rex Pugh, Cerent’s Product Line Director, explained that ATM did not strictly require fiber to work, as DS1 and DS3 transport devices could be used instead, but ATM was a fundamentally different type of transport. He says, “ATM tried to emulate voice using TDM, chopping IP up into cells for transport. However, IP was designed for statistical multiplexing, not time-division multiplexing. The precision of TDM was [simply overkill for data traffic, especially] when bandwidth in the local area network [access] was plentiful.”
That fact alone raised questions throughout the late 1990s about why a network provider would deploy ATM when it simply acts as an intermediary layer between carrying IP traffic over the needed transport systems to complete an end-to-end connection. Andy Bechtolsheim, with roots in Cisco, was intuitive when he noted in 1997, “the network technology that will win is the one that is simplest, fastest, cheapest, and easiest to use with IP.” And for emphasis, he punctuated his thoughts that the technology choice is Ethernet, not ATM.
Those sentiments were in stark contrast to one of Fore Systems’ mottoes that “All roads lead to ATM.” [1] Telecom Valley insiders, Cerent included, talked a good game about supporting ATM interfaces to move circuit-based cell traffic around the network, but ultimately this upstart startup never commercialized the capability.
Silicon Valley insiders too, especially venture capitalist luminary, Vinod Khosla, felt Fore systems, “with a price-earnings ratio of 139.13 and a market cap of $2.86 billion” in March 1997, was overvalued. Kleiner Perkins’ King Midas argued that ATM was a niche technology that would ultimately lose out to a more mainstream Ethernet/IP traffic flow, especially as faster routers arrived on the market. Khosla expected Juniper, another of his early startup investments, to deliver on that promise. Ed Kozel, Cisco’s chief technology officer at the time, declared, “ATM has stalled.”
Rex Pugh’s views resonated with Andy Bechtolsheim’s earlier 1996 predictions. Both men saw that the key to winning the IP/Ethernet battle was occurring in enterprise networks. ATM was complicated and expensive for the enterprise LAN. Business enterprises did not want to invest in the operational complexity and the costs of hiring experts for deploying and managing an ATM network. Rex notes, “Ethernet was the simpler, cheaper technology solution and it operated even faster than ATM. Ethernet chip vendors eventually won the LAN speed race over ATM.”
Business models were compared too. Whereas Ethernet supported an open, multi-vendor environment, ATM did not. The purchase of an ATM switch from one vendor meant that the user was locked into that vendor. An existing switch would not interoperate with another ATM switch from a second vendor. Ethernet, on the other hand, supported a multi-vendor environment. Concurrent with this Ethernet attribute, the SONET standard for optical transport products provided interoperability at a number of optical line rates. An optical product from one vendor would interoperate with another supplier’s product at the OC-3 rate, for example.
In addition to lacking support for multi-vendor environments, early ATM systems lacked network management support. Ethernet and optical platforms developed support for management at the outset to aid in the operator’s ability to troubleshot faults and provision data services.
Eric Cooper was steadfast during the 1990s in his belief that “ATM’s strength in the backbone is stronger than ever. And just on technical grounds, I think Gigabit Ethernet is not likely to succeed on anything but small networks.”
How wrong his prediction turned out to be, as both Gigabit Ethernet and then 10GigE took off just three years later. By March 2000, ATM switching and transport were waning technological solutions and Ethernet switching and transport (IP + Optical) were eclipsing the fading circuit-based networking in favor of packet-based connectivity.
Sig Luft provides another perspective, keeping Vinod Khosla’s early views of the terabit tsunami in mind, “100BaseT Ethernet killed 25 Mb/s ATM to the desktop; Gigabit Ethernet killed ATM on the campus; and the only place ATM could exist was in the access network. And it was that segment of the market Vinod wanted Fiberlane to address, that is, packet over SONET at the edge of the access market.”
Andy Bechtolsheim’s insight was ultimately rewarded with a single protocol—IP—delivering voice, data, and video today. Back in the 1990s, the telephone companies sought separate networks, which meant separate protocols for voice, data, and/or video. TDM accommodated voice and ATM was selected to manage data and eventually video services, both of which were designed to establish end-to-end connections in order for bit counting to take place so that customers could be billed accordingly.
Those days are long gone. The world of telecommunications transformed into one embracing Packet Optical Transport Systems, and both Cisco and Cerent had a big hand in making that happen.
[1] Eric Cooper: true believer, Alex Gove, The Red Herring, March 1997, pp. 90–92
[2] Extracts taken from Robert K. Koslowsky’s 2014 book, The Upstart Startup.
[3] Tom Nolle's post on ATM is worth reading too: How to Keep SDN/NFV From Going the Way of ATM
Those sentiments were in stark contrast to one of Fore Systems’ mottoes that “All roads lead to ATM.” [1] Telecom Valley insiders, Cerent included, talked a good game about supporting ATM interfaces to move circuit-based cell traffic around the network, but ultimately this upstart startup never commercialized the capability.
Silicon Valley insiders too, especially venture capitalist luminary, Vinod Khosla, felt Fore systems, “with a price-earnings ratio of 139.13 and a market cap of $2.86 billion” in March 1997, was overvalued. Kleiner Perkins’ King Midas argued that ATM was a niche technology that would ultimately lose out to a more mainstream Ethernet/IP traffic flow, especially as faster routers arrived on the market. Khosla expected Juniper, another of his early startup investments, to deliver on that promise. Ed Kozel, Cisco’s chief technology officer at the time, declared, “ATM has stalled.”
Rex Pugh’s views resonated with Andy Bechtolsheim’s earlier 1996 predictions. Both men saw that the key to winning the IP/Ethernet battle was occurring in enterprise networks. ATM was complicated and expensive for the enterprise LAN. Business enterprises did not want to invest in the operational complexity and the costs of hiring experts for deploying and managing an ATM network. Rex notes, “Ethernet was the simpler, cheaper technology solution and it operated even faster than ATM. Ethernet chip vendors eventually won the LAN speed race over ATM.”
Business models were compared too. Whereas Ethernet supported an open, multi-vendor environment, ATM did not. The purchase of an ATM switch from one vendor meant that the user was locked into that vendor. An existing switch would not interoperate with another ATM switch from a second vendor. Ethernet, on the other hand, supported a multi-vendor environment. Concurrent with this Ethernet attribute, the SONET standard for optical transport products provided interoperability at a number of optical line rates. An optical product from one vendor would interoperate with another supplier’s product at the OC-3 rate, for example.
In addition to lacking support for multi-vendor environments, early ATM systems lacked network management support. Ethernet and optical platforms developed support for management at the outset to aid in the operator’s ability to troubleshot faults and provision data services.
Eric Cooper was steadfast during the 1990s in his belief that “ATM’s strength in the backbone is stronger than ever. And just on technical grounds, I think Gigabit Ethernet is not likely to succeed on anything but small networks.”
How wrong his prediction turned out to be, as both Gigabit Ethernet and then 10GigE took off just three years later. By March 2000, ATM switching and transport were waning technological solutions and Ethernet switching and transport (IP + Optical) were eclipsing the fading circuit-based networking in favor of packet-based connectivity.
Sig Luft provides another perspective, keeping Vinod Khosla’s early views of the terabit tsunami in mind, “100BaseT Ethernet killed 25 Mb/s ATM to the desktop; Gigabit Ethernet killed ATM on the campus; and the only place ATM could exist was in the access network. And it was that segment of the market Vinod wanted Fiberlane to address, that is, packet over SONET at the edge of the access market.”
Andy Bechtolsheim’s insight was ultimately rewarded with a single protocol—IP—delivering voice, data, and video today. Back in the 1990s, the telephone companies sought separate networks, which meant separate protocols for voice, data, and/or video. TDM accommodated voice and ATM was selected to manage data and eventually video services, both of which were designed to establish end-to-end connections in order for bit counting to take place so that customers could be billed accordingly.
Those days are long gone. The world of telecommunications transformed into one embracing Packet Optical Transport Systems, and both Cisco and Cerent had a big hand in making that happen.
[1] Eric Cooper: true believer, Alex Gove, The Red Herring, March 1997, pp. 90–92
[2] Extracts taken from Robert K. Koslowsky’s 2014 book, The Upstart Startup.
[3] Tom Nolle's post on ATM is worth reading too: How to Keep SDN/NFV From Going the Way of ATM