Posted by: Tom Nolle
cloud architecture, data center architecture, data center networking, Juniper, QFabric
HP announced disappointing results, a contrast not only to Street expectations but to competitor Dell’s recent numbers. The problem, says the company, is softness in the consumer PC sector and the fact that HP doesn’t sell much to businesses relative to their total PC sales. The real issue, I think, is the company’s management agony and a nearly total loss of focus on business.
I was particularly unmoved by the CEO’s promise to get something going with cloud computing. Where has he been, anyway? This is no time to start laying out your cloud strategy; competitors have been doing that for over a year. HP’s decision to buy Palm is another point of future challenges; they’re not sustaining their momentum in their core markets, so how do they expect to take on Android and Apple in the tablet and smartphone space? This is a company that’s been on a roll for years, and it’s now at serious risk to lose credibility and market share. They have perhaps two quarters now to turn things around, after which they’re probably going to risk permanent damage.
The cloud also figures in the Juniper announcement. The company has been talking about their “Stratus” project for several years, and they’ve finally started delivering on the new data center fabric officially called QFabric, with the nodal element, the QFX3500. The details and the roadmap are impressive, and it’s very possible that Juniper has something here that will change the game, change some minds, and produce significant competitor angst. We’ll cover this in detail in our March Netwatcher, but let me summarize here.
The Juniper QFabric architecture consists of three elements, the primary of which are the nodes. These are essentially line cards in a basic case, designed to be linked to each other as an entry strategy or for full QFabric configurations linked back to the interconnect box. The links are made with multi-homed fiber and the result is a semi-mesh of the nodes that has a large cross-sectional bandwidth. The nodes learn the configuration and connectivity and use this to propagate a forwarding table both at Level 2 and 3, and this table then creates the full forwarding path decision so that no matter what route is taken from source to destination within the mesh, there are no further forwarding decisions needed. The configuration has a current maximum capacity of 40 Tbps and it’s fully non-blocking, lossless, has microsecond-level delay and negligible jitter.
The QFabric can be partitioned into virtual networks, and can host services that are created by attaching engines that perform the service processing. Security is an obvious example of a service. Services are created by routing data paths through the appropriate engine(s) on the way to the destination. A director device creates a black-box virtual device abstraction for the management plane and to the outside world so the structure is opaque and opex and configuration complexity are reduced.
While it’s not possible to sustain microsecond-scale latency over WAN distances, you can connect QFabric paths with a decent-performing IP/MPLS connection and thus extend the fabric beyond a single data center. This means that a cloud computing offering (either to support a service, a private cloud, or an operator IT application/feature hosting platform) could in theory be created and maintained as a single QFabric. The whole process is operationally linked vertically to the Junos Space cloud feature and management platform, and you can also use Space to create applications and service features that become services of a QFabric cloud.
What’s interesting about this beyond the obvious in-data-center benefits of cost and footprint is the notion that QFabric might become the architecture for private, public and hybrid clouds. So far, nobody has really articulated how you’d build a service provider cloud, for example, and with the WAN extensions QFabric could be just that. The capability could generate some really valuable cloud, content, and mobile engagement for Juniper and thus could pre-empt plans by competitors like Cisco to get a lead in defining how a provider cloud would look.
Since QFabric is also likely to be a compelling migration option for companies with two years or less of undepreciated data center switch assets, and at least a consideration for companies with three or even four years remaining, it could boost Juniper’s market share and credibility in the critical data center networking space. Which, obviously, neither Cisco nor HP would like. Both these arch-rivals have their own quarterly performance issues to work through, and Cisco named a COO (Gary Moore) to help them streamline its operations processes. There may be a window for Juniper to put the hurt on both companies while they’re distracted.