Denver-based Infrastructure-as-a-Service (IaaS) provider PeakColo may not have the name recognition of Amazon and Google in the marketplace, but with 300% annual growth, it has the market mojo to challenge these market leaders. PeakColo’s spectacular growth is the direct product of its localized market strategy combined with its close relationships with its key vendors – VMware, NetApp and Brocade Networks.

While the big name IaaS companies provide generic infrastructure services worldwide from a few central data centers, PeakColo works with managed service providers (MSAs) and value-added resellers (VARs) who build customized systems on top of PeakColo’s core infrastructure to meet the unique needs of the customers they have served for years and know well. While the other IaaS vendors locate their data centers to minimize cost, PeakColo’s build-out strategy is to create a distributed system with data centers in multiple tier 1 and 2 U.S. cities to minimize network latency. So far it has centers in five cities with more planned as the company continues to grow.

However, skyrocket growth also creates major challenges on the back-end, PeakColo CEO Luke Norris told the Wikibon.org community at a Peer Incite Meeting August 7, 2012. Keeping up with this level of growth while maintaining service levels, and in particular the core measurement of IO response time, which is the main measurement of service quality rather than traditional measures such as network speed, is a huge challenge.

“How do you ingest 10-20 Ethernet switches and attach thousands of compute loads to provide 2 Pbytes into an environment to be available tomorrow to meet a client’s needs while not disrupting other users on that unit?” he asks. To do that PeakColo needs a very stable infrastructure that can accommodate the addition or upgrade of components without impacting performance while boxes are switched out and replaced or added almost constantly somewhere in the network.

To create that PeakColo has virtualized its entire infrastructure using VMware and in some special instances HyperV and other hypervisors to meet the needs of individual customers. Under that, he says, over time it has run products from a list of vendors.

In the days of spinning rust, storage, and in particular disk read/write speeds, were often the gating issue in IO response. Today, as flash moves into the environment at multiple levels, the gating factor more often is the network. “We needed a network technology that could remain stable when we pull out a switch to upgrade or add new switches as we grow,” he said. The spanning tree architecture was a particular source of problems, and in the end, after investigating all the vendors, PeakColo elected to eliminate it entirely and take a gamble on the new Transparent Interconnect of Lots of Links (TRILL)  standard and specifically standardize on Brocade’s latest generation of 10 GB Ethernet TRILL switches.

PeakColo had experience with Brocade as its Fiber Channel SAN vendor. Ironically as part of the conversion to what Norris calls the company’s fifth generation infrastructure, it eliminated Fiber Channel and built a single unified network for both SAN and WAN/LAN.

This, he said, has several advantages. First Fiber Channel is losing the performance race to 10GB and 40GB Ethernet, so the new solution provides SAN speeds that at least duplicate if not exceed what the Brocade Fiber Channel switches delivered. At the same time it simplifies the network infrastructure. It means that PeakColo only buys one set of switches, manages the entire network as a single entity, and maintains one set of network management and support skills. It also simplifies interconnectivity and removes an issue in delivering high IO response levels.

It also supported a reorganization of the technical staff to focus on overall service delivery rather than on technical measures of performance of traditional IT silos. “The day when a senior tech can be completely focused on SAN throughput with no understanding of the server or application needs is over,” he says. Today PeakColo’s teams are lead by senior engineers who see the entire picture from an application standpoint and are focused on providing optimal performance of the service to the customer.

It also supports a new approach to upgrade scheduling. Upgrades are planned based on the needs of the application rather than on the lifecycle of the hardware. “If you are upgrading on a three-or five-year schedule, you are limiting your applications to that same schedule,” Norris argues. “Upgrades should be based on business needs – the growth in application demands or when the business will need that application, for instance, to start supporting Big Data analysis.”