Intel's acquisition of Ethernet chip vendor Fulcrum Microsystems is just the latest step in integrating the components within data centers to help them work smoothly as a single virtual system.
Fulcrum makes silicon for data center switches with 10-Gigabit and 40-Gigabit Ethernet ports. Its chips are known for low latency. The small, privately held company's technology will give Intel a place inside the switches atop server racks that link servers to each other and to the overall network. Intel wants to pair Fulcrum's chips with its own silicon used in Ethernet adapters and controllers so they can exchange new types of information about security, quality of service, management and other variables.
[ Keep up on the day's tech news headlines with InfoWorld's Today's Headlines: Wrap Up newsletter. ]
Intel did not disclose the price it paid for Fulcrum, which is based in Calabasas, California, and was founded in 1999. Its plan is to have most of Fulcrum's employees transfer to Intel and work in the company's data center group.
System vendors are already lining up their own computing, networking and storage platforms to help automate the operation of virtualized data centers. Cisco Systems has paired its UCS (Unified Computing System) servers with its data-center switches, along with EMC storage and VMware's virtualization software. Hewlett-Packard also sees itself offering data-center integration, and Juniper Networks and IBM seem to be heading down the same path through a partnership.
"Our customers are looking to purchase compute, networking and storage as one unit," said Steve Schultz, director of marketing at Intel. "This is a building block we didn't have in our product portfolio."
The key to these types of initiatives is to make servers and network components aware of each other so they can work more closely together, said Yankee Group analyst Zeus Kerravala. As applications on dedicated servers give way to virtual machines that can be moved around for greater efficiency, it's harder for IT administrators to keep network policies up to date manually, he said. Such an effort is both time-consuming and prone to human errors, which are the leading cause of network downtime, according to Kerravala.
No other vendor yet has all the pieces for the kind of tight integration Cisco has achieved, Kerravala said. If Intel makes server and switch chips that can talk to each other at that level, smaller manufacturers will be able to offer coordination with products from other vendors that also use Intel.
"This gives the rest of the field an opportunity to go compete with Cisco," Kerravala said. Though they wouldn't be likely to match Cisco overnight, these smaller rivals ultimately could claim high-level interoperability based on a de facto industry standard of Intel chips, he said. One customer of such an architecture might be Arista Networks, a maker of high-performance data-center switches that is already a partner of both Intel and Fulcrum. Arista's switches are geared toward very large cloud-computing infrastructures that need connections as fast as 40G bps or 100G bps. Arista was not immediately available to comment on the Intel-Fulcrum deal.
Intel has tried to enter the network infrastructure business in the past, with little success. It has a better chance to succeed now because of the closer relationship between computing and networking, Kerravala said.
Broadcom, a major rival of Intel's for both switch and controller silicon, said on Tuesday it has been able to achieve the same kinds of data-center integration capabilities in conjunction with its customers. "The silicon vendors have to provide the capabilities in their chips, and the system vendors have to then tie it all together with their software," said Rajiv Ramaswami, executive vice president and general manager of Broadcom's infrastructure and networking group.
The types of higher-level capabilities that Intel and others are talking about should cause IT administrators to look closely at interoperability, Kerravala added. While all Ethernet devices will work together at a basic level, the more advanced types of interaction aren't likely to work across all brands of gear that an enterprise may buy, potentially limiting choice.
"Even if you choose to do it the Intel way, you're still stuck with the Intel stack," Kerravala said.