The huge scale of cloud computing forces an enterprise to standardize, extend, and wring all the power out of its resources. Orchestration products like VMware are excellent at taking unused computing horsepower and putting it to work. Unfortunately, enterprise data center topology just won't let VMware get at all the horsepower.
An enterprise with rooms full of computers can't make best use of its resources to deal with its workload. For instance, if two sets of racks are running at half speed, then half the resources are wasted. Wouldn't it be better to pack that workload on one set of racks and free up the other to take on more work?
A decade of improvements
The components of the technology stack are often overhauled, to provide more capacity, instantly, at less cost. This is what has made cloud computing possible.
We've made a lot of progress in the last ten years. The cost of a megabyte of memory has dropped from 20 cents to less than a penny. The cost of shifting a megabyte down an ADSL line has dropped from $70 to a few pennies. The number of machines that can be run from one rack has gone up from 30 physical machines ten years ago to 3,000 virtual machines now. There's more of everything for everyone. And that's causing problems.
Private cloud problems for the enterprise
While some parts of the cloud revolution have raced ahead, some are stuck in the past. The new virtualization layer has been added to an enterprise's IT stack and is changing the way it does business, but the lower networking layer is pretty much the same old architecture. It is creaking under the strain of new demands, it is not controlled centrally, and it is not open to new features.
It's a problem first faced by the mobile phone industry when they moved to Internet technology. Running a secure mobile network on a rigid IP network caused network problems for carriers. Now, cloud computing is causing network problems for the enterprise.
It's not an issue for the hyperscale providers, like Amazon, Google, Facebook and Rackspace. They have created highly customized networks, so neither their problems nor their solutions apply to the rest of us. It's not an issue for the small business with four racks of computers, because the scale isn't big enough to stress the network.
An enterprise faces network problems like these.
- Adding a new server takes seconds, but configuring the network for that server takes days.
- The network team is the same size, but they must support many more network devices.
- There are many more incident tickets for network problems.
- The server landscape has become dynamic, and the network team can't keep up with policy changes.
- Network devices are struggling to keep up with all the extra decision making.
These problems may come from the way a network has grown. A common way of building new IT horsepower is to fill a single rack. More racks are built using the same one-rack design, repeated over and over. This works fine, until you try to spread workloads over several racks. Networking experts have to work hard to spread that load, and the resulting network is fragile, error-prone, and full of customized configuration to keep business flowing.
Fixing the network
The solution is to virtualize the network and provide cloud's OSSM benefits (On-demand, Self-service, Scalable and Measurable). Somehow manual processes and VLAN barriers must be removed, and controlling software must be centralized and flexible. These are the changes that will allow VMware to put all the computing horsepower to work.
SDN (Software-defined Networking) takes the decision making control off the many network computers spread around an enterprise network and puts it in a central location. The ONF (Open Networking Foundation) are promoting the OpenFlow protocol, a way of providing SDN. Many of the big switch manufacturers like Cisco, HP, Juniper and IBM have upgraded their network products to talk OpenFlow.
Big Switch Networks are a young company working to add SDN to enterprise data centers. Big Switch build commercial solutions using OpenFlow, for a client list of early adopter enterprises with between five and ten thousand servers. Big Switch have extended the SDN idea with their OpenSDN architecture, have open-sourced their Floodlight OpenFlow controller, and have encouraged a community to form around Floodlight. The Big Switch company is small, but since two founders of Big Switch and their board of directors have all had ridiculously successful careers, it won't stay that way for long.
What about the rest of us? When do we get a flexible network? The early adopters -- certain large industries that can afford to implement this new way of networking -- started discussing the changes to be made in 2011 and they are now running pilots. The rest of us -- the mainstream enterprises -- are a couple years behind. We may start talking about more flexible networks this year, run a pilot next year, and enter production in 2014.
Nick Hardiman builds and maintains the infrastructure required to run Internet services. Nick deals with the lower layers of the Internet - the machines, networks, operating systems, and applications. Nick's job stops there, and he hands over to the designers and developers who build the top layer that customers use.