Cross-connections (XCs) at the DS3 (or STS-1) level are all well and good, and the Cerent 454 had this capability from Day 1. But customers, such as Williams Communications and numerous independent telcos, also demanded DS1-level or Virtual Tributary (VT) cross-connections.
Cerent recognized this, but the upstart startup sought to introduce a product sooner rather than later and so the initial product offering hit the so-called “sweet spot” featuring DS3-fed OC-48 configurations with near-term plans to support Ethernet.
But listening to customers is the sign of a maturing startup and both Mike Hatfield and his successor, Carl Russo, listened closely. Data offerings planned for the ‘454’ were delayed in favor of the DS1 option, much to the chagrin of Vinod Khosla, Cerent’s Chairman and chief investor rooted firmly at Kleiner Perkins Caufield and Byers [1]. Besides two new DS1 modules – one working and another protection one – a second XC card would be needed.
The XC Introduced on Day 1
The first cross-connect card, known as the XC, accompanied the first release of the product. Sridhar Venkatesh led the charge on the hardware design with support form a software team that was being revamped by the newly hired software director, Gary Baldwin.
Cerent recognized this, but the upstart startup sought to introduce a product sooner rather than later and so the initial product offering hit the so-called “sweet spot” featuring DS3-fed OC-48 configurations with near-term plans to support Ethernet.
But listening to customers is the sign of a maturing startup and both Mike Hatfield and his successor, Carl Russo, listened closely. Data offerings planned for the ‘454’ were delayed in favor of the DS1 option, much to the chagrin of Vinod Khosla, Cerent’s Chairman and chief investor rooted firmly at Kleiner Perkins Caufield and Byers [1]. Besides two new DS1 modules – one working and another protection one – a second XC card would be needed.
The XC Introduced on Day 1
The first cross-connect card, known as the XC, accompanied the first release of the product. Sridhar Venkatesh led the charge on the hardware design with support form a software team that was being revamped by the newly hired software director, Gary Baldwin.
The XC card is the central element for switching traffic on the ‘454’ platform. It establishes connections and performs time division switching at the STS-1 level among all of the traffic-carrying cards [2]. This card has no external interfaces, because all of the cross-connect card’s interfaces are provided through the backplane of the ‘454.’ The XC is required for every ‘454’ system configuration.
The BTC ASIC, designed by Martin Roberts, would find a home on every one of the Cerent 454 plug-ins while the SXC ASIC, also designed by Martin, would form the heart of the XC plug-in. This hardware would be housed in the central two slots of the Cerent 454 chassis, one XC in working mode and the other XC in standby mode
Customers Shape a Startup’s Product And Drive Its Evolution
Williams and Triangle Telephone were the biggest drivers for the development of the second cross-connect card. These companies wanted the ability to manage both STS and VT circuits, but for different reasons. Williams wanted to manage both DS3 and DS1 traffic up to OC-48 initially, and later OC-192 bit rates, for its existing long distance network, while Triangle simply sought DS1 management for its OC-12-based linear network. Neither company was interested in Ethernet functionality at the time. Nor were most of the other Independent Telcos. Ethernet would take a back seat to the bandwidth tsunami, as Vinod called it, formed by the flood of TDM traffic hitting the PSTN in 1999.
Customers Shape a Startup’s Product And Drive Its Evolution
Williams and Triangle Telephone were the biggest drivers for the development of the second cross-connect card. These companies wanted the ability to manage both STS and VT circuits, but for different reasons. Williams wanted to manage both DS3 and DS1 traffic up to OC-48 initially, and later OC-192 bit rates, for its existing long distance network, while Triangle simply sought DS1 management for its OC-12-based linear network. Neither company was interested in Ethernet functionality at the time. Nor were most of the other Independent Telcos. Ethernet would take a back seat to the bandwidth tsunami, as Vinod called it, formed by the flood of TDM traffic hitting the PSTN in 1999.
So, Ajaib Bhadare, Cerent’s vice-president of engineering, marshaled his hardware and software forces and work commenced on the DS1 and VT1.5 bandwidth management options. The XCVT card became the central element for switching both DS3 (STS-1) and DS1 (VT1.5) traffic up to the OC-48 level [3]. The expanded engineering team hit the ground running.
Much of the earlier software could be reused as the STS-1 switching remained the same, but numerous VT1.5 software blocks had to be added to support concurrently developed DS1 hardware modules.
Hui Liu, Cerent’s hardware engineering director, recalls, “After the success of the re-designed BTC and SXC chips, I assigned Martin Roberts to design the VT1.5 cross-connect chip, VTXC, needed for DS1 module support.” This chip was equivalent to some sixty-four existing VT1.5 cross-connect chips, a size that would have exceeded the space allotted on Sridhar’s original cross-connect board. The VTXC chip delivered both cost and power savings on the order of 100-times. Dave [Scott] quickly added the VTXC to produce a unique cross-connect board, the XC-VT, capable of switching both VT1.5 and STS1 traffic.
Williams pushed the Cerent engineering team on the VT1.5 cross-connect capabilities. Hui adds, “Our VTXC was superior to existing chips, not just in power consumption, cost, and space, but also in our unique performing monitoring features!”
The XC-VT board became a key element in the second major release of the Cerent 454. The pace for Dave dropped down to a more manageable 60 hours a week for two reasons; one, Hui and Gary were scaling the hardware and software engineering teams by the end of 1998 and two, Gary and Ajaib decided to undertake a risky but necessary change in direction for the software of the Cerent 454. The rest of the organization was unaware that this change was about to happen.
But that is a story for another day.
[1] As I wrote in The Upstart Startup, “No one in the venture capital community wanted to fund companies who were developing DS1 implementations anymore, especially not Vinod. This low speed interface was not only boring and ho-hum, but also it flew in the face of Vinod’s vision of how to address the impending terabit tsunami. This tsunami of massive bandwidth would be tamed with new high-speed Gigabit Ethernets, not with old world low-speed DS1s.”
“Vinod was okay with DS3s on the box, but he did not want DS1s. After one Board meeting held in Petaluma, in late 1998, Vinod barged into my [R.K. Koslowsky’s] office and asked, ‘Do we really need DS1s?’”
“I calmly and firmly said, ‘Yes, we do. There won’t be any telco sales without the DS1 capability added.’”
“Vinod relented but he could not figure out why anyone would want DS1s when the terabit tsunami was about to hit. As it turned out, my assertion was validated in 2000 when the Dell’Oro Group reported that two-thirds of the interfaces sold on SONET ADMs during 1999 were DS1 speeds. The most popular DS1 application connected enterprise networks to the metro network.”
[2] The switch matrix on the XC card consists of 288 bidirectional ports. When creating bidirectional STS-1 cross-connects, each cross-connect uses two STS-1 ports. This results in 144 bidirectional STS-1 cross-connects. The switch matrix is fully crosspoint, nonblocking, and broadcast supporting. (Any STS-1 on any port can be connected to any other port, meaning that the STS cross-connections are nonblocking.) This allows network operators to concentrate or groom low-speed traffic from line cards onto high-speed transport spans and to drop low-speed traffic from transport spans onto line cards.
[3] A third cross-connect card would be needed to accommodate the larger bandwidth needs of the subsequent OC-192 option. The XC10G would arrive by 2002.
Williams pushed the Cerent engineering team on the VT1.5 cross-connect capabilities. Hui adds, “Our VTXC was superior to existing chips, not just in power consumption, cost, and space, but also in our unique performing monitoring features!”
The XC-VT board became a key element in the second major release of the Cerent 454. The pace for Dave dropped down to a more manageable 60 hours a week for two reasons; one, Hui and Gary were scaling the hardware and software engineering teams by the end of 1998 and two, Gary and Ajaib decided to undertake a risky but necessary change in direction for the software of the Cerent 454. The rest of the organization was unaware that this change was about to happen.
But that is a story for another day.
[1] As I wrote in The Upstart Startup, “No one in the venture capital community wanted to fund companies who were developing DS1 implementations anymore, especially not Vinod. This low speed interface was not only boring and ho-hum, but also it flew in the face of Vinod’s vision of how to address the impending terabit tsunami. This tsunami of massive bandwidth would be tamed with new high-speed Gigabit Ethernets, not with old world low-speed DS1s.”
“Vinod was okay with DS3s on the box, but he did not want DS1s. After one Board meeting held in Petaluma, in late 1998, Vinod barged into my [R.K. Koslowsky’s] office and asked, ‘Do we really need DS1s?’”
“I calmly and firmly said, ‘Yes, we do. There won’t be any telco sales without the DS1 capability added.’”
“Vinod relented but he could not figure out why anyone would want DS1s when the terabit tsunami was about to hit. As it turned out, my assertion was validated in 2000 when the Dell’Oro Group reported that two-thirds of the interfaces sold on SONET ADMs during 1999 were DS1 speeds. The most popular DS1 application connected enterprise networks to the metro network.”
[2] The switch matrix on the XC card consists of 288 bidirectional ports. When creating bidirectional STS-1 cross-connects, each cross-connect uses two STS-1 ports. This results in 144 bidirectional STS-1 cross-connects. The switch matrix is fully crosspoint, nonblocking, and broadcast supporting. (Any STS-1 on any port can be connected to any other port, meaning that the STS cross-connections are nonblocking.) This allows network operators to concentrate or groom low-speed traffic from line cards onto high-speed transport spans and to drop low-speed traffic from transport spans onto line cards.
[3] A third cross-connect card would be needed to accommodate the larger bandwidth needs of the subsequent OC-192 option. The XC10G would arrive by 2002.