Juniper Networks (s jnpr) unveiled a new data center architecture that will flatten out the network layer to improve performance, scale and manageability. Called QFabric (short for Quantum Fabric), the new product line will help customers to create a fabric of servers, storage and networking from their data center resources, all of which can be managed from a single location, and all of which connect to a series of top-of-rack nodes that acts as a single logical switch. It’s a bold move for Juniper to start pushing a new network architecture, but the company claims QFabric was designed to handle the increasing demands put on the network from mobile applications and cloud computing, so the timing is right.
Essentially, the rationale behind QFabric is that data center networks are under increasing pressure to transport lots of data, carry out many transactions and deliver more services, and that the traditional multi-tier network architecture has become the bottleneck. Cisco and others — including Juniper — have built their data center businesses on this model, but Juniper has spent three years creating QFabric, under the Project Condor codename, to change this approach.
As explained by Juniper CTO and Founder Pradeep Sindhu during the event, QFabric consists of three primary components: nodes, interconnects and director software. The nodes provide the business logic and silicon that forward packets and operate the QFabric control plane; the interconnects are essentially edge components full of fabric cards that transport data between network devices; and the director software provides the management interface full the entirety of a QFabric network. According to Juniper, there is a maximum of one hop between nodes in a QFabric architecture, and maximum cross-data-center latency between nodes of 5 microseconds. One test customer, the National Energy Scientific Computing Center, found that QFabric’s Ethernet-based approach approached and, in some cases, outperformed an InfiniBand cluster in certain benchmarks.
The first QFabric product, the QFX3500, will be available this quarter, and can act as an Ethernet switch for the time being before transitioning to a QFabric node when the rest of the product line becomes available in the third quarter.
Of course, the elephant in the room when talking about anything network is Cisco (s csco), which still dominates that switch market despite decreasing switch revenue. I spoke with Juniper CEO Kevin Johnson after the press event, who acknowledged that it will take a while before Juniper — which currently accounts for 2.3 percent of the switch market — can make a serious dent in Cisco’s market share, or even to contribute significantly to Juniper’s bottom line. However, he noted that concerns over network performance and complexity certainly will be even greater when new hardware replacement cycles kick in five to seven years from now, which makes QFabric very promising in the long run. I asked whether that gives too much time for competitors to follow suit, but Johnson thinks the amount of time and resources Juniper put into QFabric — 1 million-plus man hours and more than $100 million — make it difficult to replicate, especially because Juniper will keep moving forward on it.
And QFabric does have one somewhat critical limitation right now, which is that it’s constrained to a single data center deployment. However, in response to question from the live audience during the event, Juniper’s David Yen noted that the company is working on connecting QFabric data centers across the WAN to help create multi-node cloud platforms, among other things.
At this point in the QFabric story, of course, everything sounds great, but there is a lot that can happen. From competitors such as Cisco, Arista or Brocade (s brcd) making inroads — all actually are pushing fabric architectures of one shape or another — to customers simply not buying in, QFabric is by no means the next big thing in networks just yet. Whatever the result, though, it seems logical that the vision behind QFabric will ultimately help reshape the data center network, regardless who’s providing the gear.
Related GigaOM Pro research (sub req’d):
- For Open Cloud Computing Look Inside Your Data Center
- Defining Internal Cloud Options: From Appistry to VMware
- If We Compute in the Cloud, We Need a Network Fabric