bluebay2014 - Fotolia
Fiber Mountain wants you to dump your data center core and aggregation switches and replace them with bunches of intelligent fiber strands, connected to top-of-rack switches and an SDN controller.
The company came out of stealth mode at Interop NYC last week with a data center architecture called Glass Core, which uses an optical fabric to provide flat, any-to-any connectivity. Fiber Mountain claims that a non-blocked, fiber-based architecture will reduce latency, increase throughput and slash spending on hardware and power.
Traditional network hardware vendors love the "feed the core" philosophy, said Fiber Mountain CEO M.H. Raza, who led development on the company's optical fabric. Fiber Mountain intends to end the "proliferation of layers of switches."
The concept of dumping expensive hardware for cheaper optics was clearly attractive at Interop. On an otherwise sleepy show floor, Fiber Mountain's booth had lines of visitors, and Raza gave one Glass Core demonstration after another.
But Fiber Mountain has plenty more evangelizing to do.
"Their biggest challenge is to get customers to acknowledge there is the possibility of improving on the performance, latency and capacity of networking … without relying on the tried-and-tested two- or three-tier infrastructure," said Rohit Mehra, vice president of network infrastructure at analyst firm IDC. "This is not always in the mind-set of the typical network manager."
How Glass Core works
The Glass Core architecture consists of a centralized controller, called the Alpine Orchestration System (AOS), as well as Optical Edge ToR switches and intelligent strands of interconnected fiber. The fiber is fed into the edge switches at up to 1,000 strands to a single rack unit space.
M.H. RazaCEO, Fiber Mountain
The AOS controller discovers the entire network topology by reading each intelligent fiber strand. Using this information, it pushes out policy and configuration to ToR switches that process the packets. AOS dynamically provisions Programmable Light Paths between any two points at 10, 40 or 100 Gbps across the core, automatically moving or changing connections when necessary. The basic idea is to create virtual tunnels over fiber between every point on the network so that traffic never has to bounce to multiple tiers of switches.
Raza claims this architecture provides up to two times the capacity of a traditional architecture, at half to one-third of the cost.
He also says moving to Fiber Mountain's Glass Core won't require a data center infrastructure rip-and-replace.
"Give us one row. We will coexist with your existing vendor [architecture]. That one row will be half the cost and twice the capacity," said Raza. Each transitioned row will connect directly to the core, and as rows are added they'll connect to each other, he explained.
Engineers will probably start with just one rack for proof-of-concept testing on non-critical applications, Mehra said. Once Fiber Mountain is in the customer's network, advanced throughput will be more of a selling point than cost, at least initially.
If Fiber Mountain's technology lives up to initial marketing claims, hardware vendors could see a real threat. Raza says Fiber Mountain is in talks with several large data center customers. The problem won't be attracting customer attention, but rather scaling to meet demand that could come rapidly, he added. The company is venture-backed and still running on a tight budget.
But the Fiber Mountain team is made up of "seasoned technology stalwarts, who know how to run a technology business," said Mehra. Raza held senior roles at ADC Telecommunications, 3Com, Fujitsu and General DataComm, and some of his current team are executives he worked with in those organizations.
In these initial months, Fiber Mountain's first five or 10 customers will "make or break" the company, Mehra said.
Is there a future in SDN and optical transport networks?
Multi-layer SDN lets operators manage across traditional and optical networks
With Plexxi SDN, no need for a fabric