NNArchive
Juniper Employee , Juniper Employee Juniper Employee
NNArchive
Spin the bottle and kiss that switch….
Nov 16, 2011

bottlesmall.jpgSwitches today are like a sports car with no doors:  while you get most of what you want, you can’t have it all.  With switches, it seems that every new feature in one area imposes a penalty somewhere else. For instance:

Layer 2 only:  Blazingly fast, but no Layer 3 features.
Layer 2 + Layer 3:  Nice feature mix, but a slight performance pause—which frames am I switching?
Layer 2 + Layer 3 + FCoE: Even more features, but the switch gets dizzy thinking about which frame is which…
Layer 2 + Layer 3 + FCoE + FC: Lots of features, but I might as well walk the frame to its destination…

 

Is it possible to get all the above without compromising in features and performance? 

 

Why is it that customers still have to choose among feature sets when buying a switch?  Better yet, why is there a lack of choice in the market today for top-of-rack access switches?  Let’s take a look at what’s currently available.

 

How about 48 ports of 10GbE, four 40 Gbps ports, L2, L3, DCB and FCoE, all in a one rack-unit top-of-rack (ToR) switch.  Sound familiar?  It should; if you’re looking for a 10GbE ToR switch to connect servers in your data center, you’ve probably noticed that most vendors offer some version of this device.  One may be blue, another may be gray and a third may be purple, but aside from some subtle differences, they are all pretty much the same. 

 

Cisco, Arista, Force10/Dell, IBM BNT—peel back the covers on any of these vendors’ switches and you’ll probably see one of two things:   

 

1) Switch manufacturers are simply taking all the same ASICs that silicon vendors are selling them, slapping them into a chassis, and pasting their logos on the box.  The only real semblance of customization is the fact that these manufacturers use their own software to enable the features the merchant silicon vendors built onto their ASICs.  And that’s fine, since without the custom software, all these ToR switches would end up looking, smelling and performing the same. 

 

2) Vendors who actually do use their own home-grown ASICs and software to power their ToR switches are forced to make compromises between performance and features.  While this does help to differentiate these products from the competition, as mentioned previously, there is always a tradeoff.  Will the switch be fast, which means Layer 2 only, or will it be feature rich (L2+L3+the kitchen sink) and..well…not so fast? 

 

Either way, when selecting a switch, you might as well close your eyes, spin the bottle and pucker up, because it really doesn’t matter which product it lands on.

 

It wasn’t always this way.  In the past, network vendors built their own ASICs and added their own software to enable their own built-in features—the key differentiators that made the switches stand out.  Now, this trend towards one-size-fits-all is raising legitimate accusations of commoditization (a topic I will address in a future blog).

 

I’ve spent the better part of my career as an ASIC designer.  I helped build key features into the top network vendor’s silicon.  And frankly, merchant silicon vendors have caught up to the networking vendors and are now producing some really good ASICs, making it easy for the manufacturers to spin out full-featured, high-quality top of rack switches.

 

So the real question customers need to ask is, what are networking vendors doing with this merchant silicon to avoid the “commoditization” label and set themselves apart?  What extra value are they building into their switches to differentiate themselves from the competition?  Why should customers buy one 10GbE ToR switch over another, aside from color preferences?  The bottle is still spinning…..

 

bottle.jpg

 

   

Juniper’s QFX3500 is an example of how network vendors should be building their 10GbE ToR switches.  The QFX3500 offers all the same features as the Cisco, Arista, Force10/Dell and IBM BNT ToR switches, with one big difference:  the QFX3500 is also an FCoE-to-FC gateway, enabling customers to connect their Fibre Channel storage network to their 10GbE network.  The QFX3500 is flexible enough to satisfy virtually all of a customer’s ToR convergence needs, whether to support a pure FCoE environment, bridge a customer’s Ethernet network to their FC network, or just connect a POD of servers.  If customers don’t need these convergence features today, they can simply turn them off; as the business grows in the future, those features are still there and can be enabled as needed.  This is called “investment protection”—something the rest of the industry seems to be ignoring.  Apparently it’s easier to change a switch’s color than spend time designing extra software features that add value for the customer.

 

 

 

 

 

One of the hallmarks of Juniper’s product portfolio is that features are shared across multiple product lines through a common operating system: Junos. The QFX3500 is no exception; it also runs Junos—by far one of the most stable networking operating system out in the market—and therefore leverages features written for other Juniper products, ensuring consistency across product families.

 

So let’s say the spinning bottle stops at the QFX3500. What would a customer get? They would get a line-rate Layer 2 and Layer 3 ToR device that includes storage convergence features—without sacrificing performance. The QFX3500 uses standards-based protocols like Layer 2, Layer 3, DCB, FCoE and FC to connect to existing networking gear in the data center, so customers can plug in and start running these features today. And it does it all blazingly fast, with latency under 1us.

 

Lastly, the QFX3500 protects customers’ investments by allowing them to reuse the device as a high-speed edge device in a QFabric system. In QFabric mode—achieved through a simple software upgrade—the QFX3500 allows customers to scale from 48 to 6,144 10GbE ports of 10GbE in a single “switch,” with latency of less than 5us from ingress port to egress port.

 

A reasonable question to ask at this point is, how does the QFX3500 help a customer scale to 6,144 10GbE ports with latency of under 5us? Actually it’s pretty simple: it’s because the QFabric architecture is radically different from anything the competition has to offer.

 

The QFabric architecture is based on a simplified core and feature-rich edge, and the QFX3500 is the fast, feature-rich edge device that enables customers to scale out their network. Customers can start with one QFX3500 (48 ports); as they grow to a couple hundred ports (four or more QFX3500s), they can transition to a QFabric system, flattening their network to a single tier and managing all their QFX3500s (called QFabric Nodes in this application) as one device.

 

While other network vendors are content to build the same switch as the competition, with no real standout features outside of what merchant silicon vendors have sold them, Juniper has taken the market to a new level by innovating and building a 10GbE ToR switch that can be used as an FCoE-to-FC gateway and also migrate to a flat one-tier fabric. All powered by Junos, one of the most stable, feature-rich operating system in the industry. So customers do have a choice: they can spin the bottle and be satisfied with whatever vanilla 10GbE ToR switch it lands on, or they can take off the blindfold and pick a truly innovative and unique solution like the QFX3500.