SAN JOSE – Broadcom brings to Interop 2012 next week another strategic piece of the puzzle for the 100 Gbit/second networks that are just starting to emerge. The BCM88650 is a high-end switching fabric that sports two 100G ports and also can be flexibly used for 10 and 40G nets.
The 40 nm switch is the first to fuse the company’s serdes and packet processing capabilities with the high-end switching fabric Broadcom acquired with Dune Networks. The chip aims to displace ASICs network systems makers design for a variety of their telecom and data center systems.
The switch includes traffic management, packet processing and media access control functions. The chip can be configured to support about 24 10G, six 40G or two 100G ports.
The chip targets a range of core, aggregation and top-of-rack switches. It also can be used as a fabric tying together as many as 4,000 100G ports across multiple cards and systems.
“We will reset the market in density and cost for 100G, and we think this will start a new cycle of 100G adoption to make the technology more mainstream,” said Nick Ilyadis, chief technology officer of Broadcom’s infrastructure and networking group.
The news comes just a week after Broadcom announced a 100G network processor, another piece of the puzzle for 100G networks just starting to roll into the market. Currently 10 and 1G switches represent the majority of today’s shipments, and 10G is growing to become the largest slice of the pie.
The 100G links are just starting to be deployed in telecom and high-end data center networks. The Broadcom chip can be used for any of the 10, 40 or 100G links.
The chip essentially doubles the throughput of Broadcom’s previous high-end switch. It likely will find most of its use handling traffic at 10 and 40G, said Jag Bolaria, a senior analyst at the Linley Group (Mountain View, Calif.).
“Most of the designs today are at 10G--that’s where the bulk of shipments are--so if Broadcom relied on 100G sockets for this product, they would not have a strong return on investment,” Bolaria said.
Indeed, Linley Group estimates merchant chips shipping in 2015 will power only about 100,000 ports of 100G links. Shipments of merchant chips for 40G ports will be triple that size, and both will be dwarfed by the dominant 10G market, it predicts.
@markhahn: it is probably because IB has not been successful in penetrating the datacenter market and remains a niche for HPC clusters. Secondly, switches with CXP/QSFP interconnects have fabric extenders that support downstream 10Gig RJ-45, a great advantage for legacy installations that IB doesn't serve.
hah, very droll! at two-port switch indeed!
can the switch actually be configured to any use whose bandwidth sums to less than 240 Gb? such as 1x 100Gb and 14 10Gb?
I personally have a hard time understanding the lassitude of the eth market - for those who actually care about latency and bandwidth, the IB world has been delivering 36-port 40 or 56Gb fabrics for quite a while. in what way does a mere 240Gb device "reset the market"?
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.