One of Juniper’s competitors is pointing at QFabric and screaming PROPRIETARY! in the hope that you’ll run away. In the name of all publicity is good publicity, we’re flattered. In the name of understanding how a QFabric switch can scale in or out of a data center environment, I'll take the bait.
The most important function of any network device is interoperability with whatever precedes it. Like most network devices QFabric has some proprietary internal components, but no proprietary external interfaces. Today a number of QFabrics are connected to legacy network devices, servers, and storage devices the same way any chassis switch would connect those devices: with Ethernet. QFabrics are switches, not networks. The distinction is important.
Let’s start with a deeper look at the word "Proprietary"
Proprietary is a dirty word in networking, and for good reason. Wikipedia defines a proprietary protocol as one controlled by a single company. Customers have seen the damage wrought by proprietary protocols that limit network investments, leading to vendor lock-ins or forcing unnatural technology adoptions and economic behavior. Vendors fling the “P” word at each other hoping customers will be too lazy to investigate. Sometimes it’s justified, sometimes not. How can you tell?
Throughout the history of networking, IT proprietary protocols and interfaces can be grouped into three main categories:
Type 1: Proprietary Networks: The worst kind of proprietary: a network composed of devices from a single vendor who is also the only company that supports a particular protocol. The only way to add or extend the network is to add gear from that same vendor. Extracting oneself from a proprietary network typically involves significant changes - and this is the important bit - to the ENTIRE network.
Examples include:Cisco EIGRP, DecNet, and legacy TDM phone systems from Avaya, Nortel, Siemens - the list goes on…
What to ask: If I want to curtail or remove network investment X, what happens?
Type 2: “Pre-Standard” Proprietary Protocols: A potentially nefarious, sometimes benign version. In version Type 2 network, a vendor touts an innovative way of solving a problem while pointing to a pending standard working its way through standards bodies. In this situation, the customer has to make a judgment on whether or not the vendor has the proper incentives to smoothly migrate to the final standard when it arrives. Without proper incentives, “pre-standard” or its cousin “extensions to the standard” can turn what looked like a bleeding edge feature into a Type 1 network (see above definition).
Examples include: PoE+, Cisco FabricPath/Brocade VCS & Trill, Cisco Skinny/SIP, 802.11n.
What to ask: How quickly will the protocol become a standard? What motivations and history does the vendor have in migrating from pre-standards to final standards?
Type 3: Embedded Proprietary: Largely benign, occasionally annoying, sometimes carries an economic incentive. In this version, a technology vendor does something inside a device that solves a problem, but does not inhibit the customer’s ability to migrate elsewhere or interoperate with network devices built by competitive vendors. While a proprietary protocol or interface may limit customers to buying certain components for the device from a single vendor, the device as a whole supports the industry-standard interfaces needed to curtail investments in the device or vendor, should that need arise. Power supplies are a common offender, as any cell phone user can attest.
Examples include:Ethernet chassis switch line card interfaces, power supplies, WLAN access points and controllers, and laptop and cell phone chargers.
What to ask: Do the benefits of the system outweigh the initial and ongoing economic costs?
Is QFabric Proprietary?
Fact: A QFabric switch is no more or less proprietary than any Ethernet chassis switch on the market today.
A “Monster truck-sized” Ethernet Chassis*
All Ethernet chassis switches are systems built of (Type 3) proprietary parts, but they connect to other network devices, servers, and storage with industry standard interfaces - mostly Ethernet. A QFabric switch behaves in exactly the same way. What makes it different is the form factor - a QFabric switch is made up of three components that historically only existed INSIDE the sheet metal of a chassis. By placing those three components into separate housings - the sheet metal and the scaling model changes, but the ability to connect to existing or new devices is exactly the same.
Since the components that make up a QFabric switch are no longer housed in the same sheet metal, the first reaction many have is to treat the components like a network, instead of as a single switch system which is wrong. However, just like a chassis switch, the internals of the system are embedded (Type 3) and the external interfaces are industry standard interfaces.
How do I curtail or grow an Ethernet chassis or QFabric investment?
When I invest in an Ethernet chassis system, I begin by buying a chassis and a few line cards. Then I add line cards until I reach the limitations of the system - typically 8 to 16 line cards. To scale beyond the capacity of a chassis, I add more network devices (complexity, latency, and risk). If needed, I can curtail my investment in that chassis at any point and connect new network devices to the legacy line cards.
For QFabric, I can do the same thing, although my buying pattern may be different - today, I begin by buying “line cards” that look and act like blazingly fast 48 port 10 Gigabit top-of-rack switches. When the traffic from the first few “line cards” begins to over run the legacy network core, I add the rest of the system. Then I stop adding capacity by buying the line cards when I either want to curtail the investment (for a specific reason), or when I hit the limitations of the system at 128 line cards. If I need more than 6,144 10bE ports, I can expand the network beyond the one QFabric system and add another QFabric or Ethernet chassis; they would use the same protocols and interfaces to connect. The economics and scaling properties of these two switch systems are different, but from the view of what is proprietary and what is not, they are the same.
Building Networks Differently
What makes QFabric so unique from other solutions targeted at distributed applications in a datacenter? In short, it’s the difference between building a new kind of switch fabric from the ground up as opposed to building protocol fabrics via layering new protocols (Type 1? Type 2?) on top of traditional network devices.
QFabric switches and Juniper’s highly successful EX Series switches with Virtual Chassis technology are switch fabrics built from the ground up to solve the new needs of distributed applications and virtualized datacenters. These technologies reduce managed devices and optimize distributed application performance without asking customers to learn, manage, and troubleshoot a new protocol. Three thousand customers in three years is a testament to the flexibility and simplicity the Virtual Chassis technology brings to both datacenter and campus networks, while QFabric offers an architecture for next generation data centers.
The bottom line:
For a QFabric solution to achieve the unparalleled economics, performance and simplicity that it does at a large scale (and with resiliency), certain internal processes happen differently than traditional network devices, and being different can sometimes mean being misunderstood. QFabric interoperates with what exists today and offers customers an alternative solution to evolve their data centers in a way that offers flexibility and scale. You can start with a small QFabric in a corner of the data center and then scale it down if you don’t like it, just as easily as you could scale it up. Next question?
*Thanks to Greg Ferro at Etherealmind.com for this inspiration.
Exploring the vision for the networking industry and the issues shaping its future.