Cloud computing has become a reality, yet the hype surrounding cloud has started to exceed the laws of physics and economics.  The robust cloud (of all software on demand that will replace the enterprise data center) will crash into some of the same barriers and diseconomies that are facing enterprise IT today.

Certainly there will always be a business case for elements of cloud, from Google’s pre-enterprise applications to Amazon’s popular services and the powerhouse of CRM, HR and other popular cloud services.  Yet there are substantial economic barriers to entry based on the nature of today’s static infrastructure.

We’ve seen this collision between new software demands and network infrastructure many times before, as it has powered generations of innovation around TCP/IP, network security and traffic management and optimization.

It has produced a lineup of successful public companies well positioned to lead the next tech boom, which may even be recession-proof.  Cisco, F5 Networks, Riverbed and even VMware promise to benefit from this new infrastructure and the level of connectivity intelligence it promises.  (More about these companies and others later in this article.)

Static Infrastructure meets Dynamic Systems and Endpoints

I recently wrote about clouds, networks and recessions by taking a macro perspective on the evolution of the network and a coming likely recession.  I also cited virtualization security as an example of yet another big bounce between more robust systems and static infrastructure that has slowed technology adoption and created demands for newer and more sophisticated solutions.

I posited that VMware was a victim of expectations enabled by the promise of the virtualized data center muted with technological limitations its technology partners could not address quickly enough.  Clearly the network infrastructure has to evolve to the next level and enable new economies of scale.  And I think it will.

Until the current network evolves into a more dynamic infrastructure, all bets are off on the payoffs of pretty much every major IT initiative on the horizon today, including cost-cutting measures that would be employed in order to shrink operating costs without shrinking the network.

Automation and control has been both a key driver and a barrier for the adoption of new technology as well as an enterprise’s ability to monetize past investments.  Increasingly complex networks are requiring escalating rates of manual intervention.  This dynamic will have more impact on IT spending over the next five years than the global recession, because automation is often the best answer to the productivity and expense challenge.

Networks Frequently Start with Reliance on Manual Labor

The Switchboard Symbolizes an Era

A very similar scenario is playing itself out in the TCP/IP network as enterprise networks grow in size and complexity and begin handling traffic in between more dynamic systems and endpoints.  The recent Computerworld survey (sponsored by Infoblox) shows larger networks paying a higher IPAM price per IP address than smaller networks.  As I mentioned, this shows clear evidence of networks growing into diseconomies of scale.

Acting on a hunch, I asked Computerworld to pull more data based on network size, and they were able to break their findings down into 3 network size categories: 1) under 1000 IP addresses; 2) 1k-10k IP addresses; and 3) more than 10k IP addresses.  Because the survey was only based on about 200 interviews I couldn’t break the trends down any farther without taking some statistical leaps with small samples.

IPAM Costs Rise as Networks Grow

In a 2.0 world servers are virtual and dynamic, and move around even more frequently than wireless laptops and phones.  While the DHCP protocol can assign addresses dynamically - and lots of other configuration data too (like the address for critical infrastructure elements like the network gateway router, the DNS server, even device-specific configuration info, etc.), the pools of addresses handed out by DHCP have to be managed, and there are lots of reasons why admins need to know which device received a particular address - and applications need to able to reach devices by name (e.g. Windows host name) versus an IP address. 

Perhaps it takes 30 minutes on average to find an address, allocate it, get a device configured, update the spreadsheet and update DNS.  That was more manageable in a static world, though the increasing cost/IP to perform these tasks in larger networks is a direct consequence of manual systems breaking down in the face of scale.  Now consider a 30 minute process for a device - or a virtual application instance - that changes IPs every few hours, or faster.  When a 1.0 infrastructure meets 2.0 requirements, things start to break pretty quickly.

That is why, even with the simple act of managing an enterprise network’s IP addresses, which is critical to the availability and proper functioning of the network, cost actually goes up as IP addresses are added.  As TCP/IP continues to spread and take productivity to new heights, management costs are already escalating. 

This is a very fundamental observation based on one of the most common network management tasks. You can assume that there are other slopes even steeper because of complexity and reliance upon manual labor.

Some enterprises are already paying even higher expenses per IP address, and chances are they don’t even know it because these expenses are being hidden within network operations.  Reducing headcounts risk increasing these costs further or making substantial sacrifices in network availability and flexibility.

IPAM as the Switchboard Metaphor

If something as simple and straightforward as IP address management doesn’t scale, imagine the impacts of more complex network management tasks, like those involved with consolidation, compliance, security, and virtualization.  There are probably many other opportunities for automation tucked away within many IT departments in the mesh between static infrastructure and moving, dynamic systems and endpoints.

This will  force enterprise IT departments into similar discussions as those which likely took place decades ago within the Bell System when telecom executives looked at the dramatic increase in the use and distribution of telephones and mushrooming requirements for operators and switchboards and offices and salaries and benefits.  One can only imagine the costs and challenges that we would face today if the basic connection decisions were still made by a human operator.

The counterpart to the switchboard of yesteryear for IPAM is the spreadsheet of today.  Networking pros in most enterprises manage IP addresses using “freeware” that has an ugly underside; it produces escalating hidden expenses that are only now being recognized, mostly by large enterprises.  Mix the growth of the network with new dynamic applications and new factors of mobility with a little human error and you have a recipe for availability, security and TCO issues.

Many of these switchboards can probably be bought or manufactured today for a song, yet it is the other costs (TCO and availability and flexibility) which make them cost-prohibitive.

Server Déjà vu

Another one of the TCO fables that are similarly bound to take the steam out of cloud fantasies has to do with hardware expenses.  The cloudplex will utilize racks of commodity servers populated with VMs that can scale up as needed in order to save electricity and make IT more flexible.  That makes incredibly good sense, but are we really there yet?  No.

Servers have a very large manual labor component, according to an IDC Report hosted at Microsoft.com.  The drumbeat for real estate and electricity savings may play well to the bigger picture buyer; yet perhaps the real payoff of virtualization is its potential to automate manual tasks, like creating and moving a server on demand.

Manual Labor is Also a Server TCO Issue

Yet one of the core promises of virtualization is to automate the deployment of server power.  If this is constrained by infrastructure1.0 (as I’m suggesting) then VMware and its partners need to address the “static infrastructure meets dynamic processing power” challenge rapidly in order to achieve levels of growth once expected in 2007.  With Microsoft now in the virtualization market thanks to Hyper-V, VMware’s window of first mover advantage is starting to close.

Virtualization security now risks becoming a metaphor for other technology-related issues that could slow down the adoption of virtualization in the lucrative production data center market. 

Netsec Wasn’t Ready for Virtsec 

The lack of network security connectivity intelligence meant that security policy, for example, would limit VMotion to within hardware-centric hypervisor VLANs.  Network security infrastructure wasn’t prepared for the challenges of protecting moving, state-changing servers, despite the promise of a stellar lineup of VMSafe partners.

The promise of virtualization that drove VMware’s stock price into the clouds eventually met up with lowered growth expectations as deployments were impacted by the lack of connectivity intelligence that no doubt impacted other potential business cases for the unquestionable power of virtualization to someday unleash new economies of scale and computing power.  These issues too will hit the cloud dream as they have also impacted other initiatives, albeit on a smaller and less visible scale.

Today there are plenty of new initiatives facing mounting pressures for connectivity intelligence and automation that have already left enterprise CIOs holding the bag for similar ecosystem finger-pointing.  Whether or not we enter a global recession, these pressures will continue and likely worsen.  They are artifacts of years of application, network and endpoint intelligence promises colliding with static TCP/IP infrastructure. 

Saving money by cutting network operations or capital budgets is the equivalent of Ma Bell laying off operators or closing switchboards in the midst of unstoppable growth.  Automation is the only way out, as Cisco’s Chambers hinted recently.

Connectivity Intelligence is Strategic to I2.0

As much as cloud computing has rallied behind the prospect of electricity and real estate savings, the business case still feels like a dotcom hangover in some cases.  Virtualization is still a bit hamstrung in the enterprise by the disconnect between static infrastructure and moving, state-changing VMs; and labor is the largest cost component of server TCO (IDC findings) and a significant component of network TCO (as suggested by the Computerworld findings).  So just how much will real estate and electricity savings offset other diseconomies and barriers in the cloud game?  I think cloud computing will also have to innovate in areas like automation and connectivity intelligence. 

For the network to be dynamic, for example, it needs continuous, dynamic connectivity at the core network services level.  Network, endpoint and application intelligence will all depend upon connectivity intelligence in order to evolve into dynamic, automated systems that don’t require escalating manual intervention in the face of network expansion and rising system and endpoint demands.

Getting beyond Infrastructure1.0’s Zero Sum Game