Until now, QFabric was targeted at financial companies, cloud providers and other organizations with very large-scale data centers that handle several hundred servers and thousands of ports. But that approach made it difficult to amass a large number of customers, which has gained Juniper lots of criticism despite all of the praise its technology receives.
While the original Juniper Qfabric, now called the QFX3000-G, scaled up to handle 6,144 10 GbE ports, the new mid-market QFX3000-M aims to address data centers with a few hundred servers and up to 768 10 GbE ports.
"The problem with the original QFabric solution was that it was so big and the initial investment was so large, that it was difficult to get started," said Gartner Inc. vice president Mark Fabbi. "This will at least get product out to the market, so we'll see more use cases and see how well it works."
Juniper QFabric upgrades: An ecosystem of products
As part of this week's announcement, Juniper launched an extension of its Virtual Chassis technology in the EX8200 Ethernet core switch line. Now up to eight switches can be managed as one logical device across four locations as far apart as 80 kilometers. Juniper also released a $40,000, 40 GbE top-of-rack switch, the QFX3600, which has16 40G ports in a 1 RU footprint. But the piece most essential to the scaled-down QFabric is the smaller QFabric Interconnect technology -- the QFX3600-I.
With QFabric, Juniper users can create large, flat Layer 2 or Layer 3 data-center networks that provide any-to-any connectivity between nodes, making each one equidistant and enabling every switch in the system to be managed as one device. To do that, QFabric has three elements: 1) a top-of-rack switch node; 2) the QFabric Interconnect where all access switches meet; and 3) the QFabric Director, which externalizes the network control plane, centralizing management.
More on data center network fabric
With edge software overlays, is network fabric just for raw bandwidth?
Wading through the data center fabric and FCoE hype
The new QFX3600-I can be used with the original QFabric node and Director but costs about one-third of the previous Interconnect, is 54% less power hungry, is one-eighth the size and claims 3 microseconds from server to server instead of 5 microseconds, said Dhritiman Dasgupta, Juniper senior director of product marketing.
Juniper QFabric: Helping customers migrate slowly
To really get the benefit of QFabric, customers must implement all three elements of the system, which is a major investment even with the smaller configuration. Until now, Juniper urged some customers to migrate slowly by first buying the top-of-rack switch and later fully altering network design by implementing the Interconnect and Director.
With this new configuration, Juniper still expects a slow migration. Dasgupta says that even some organizations with large-scale data centers will start with a small QFabric configuration and grow down the road.
"They can start with QFX3000-M and once they hit 768 ports, they can swap out the fixed Interconnect with the chassis Interconnect," he said.
Is Juniper QFabric too late to smash the network fabric competition?
Juniper is marketing the QFabric QFX3000-M as direct competition to the Cisco Nexus 7000 and 2000 core/access data center combination. Dasgupta says QFabric offers a 40 GbE fabric as opposed to Cisco's 10 GbE fabric. He also claims it is four times faster, requires far fewer devices, less power, fewer cables and can be managed as one device, unlike the Nexus switches.
Analysts agree that Juniper's technology is more innovative than many competitors, offering lower latency and smoother design, but the company still has an uphill battle.
"QFabric is a step back to the old high-performance time domain multiplexers from eons back … only updated for the new century," said research director Eric Hanselman of 451 Research. The system offers super low and deterministic latency across the backplane, which is groundbreaking. Nevertheless, "the enterprise market is going to be harder to crack because it's an architecture that's different."
In addition, Juniper's slow time-to-market gave competitors the opportunity to make progress in reducing latency.
"Some [vendors with] traditional approaches have already caught up. Alcatel-Lucent has a standard switching platform that has driven down latency," Fabbi said. "If they had come up with this when they introduced QFabric, they could have ramped up traction much more quickly."
What's more, while Juniper's data center fabricmay be more efficient than the Cisco Nexus 7000/5000 combination, the company will have a tough time taking that message to market. For one thing, its partner channel is dwarfed by Cisco's. In addition, there are fewer opportunities to partner with large vendors for OEM deals. Many once thought that both Dell would primarily sell Juniper equipment, but now Dell has acquired Force10. Meanwhile, though IBM sells QFabric, it has also partnered with NEC to sell OpenFlow-based networking, Fabbi said.