The Case for Converged Infrastructure
IT organizations are straining to survive the pressures of keeping up with explosive data growth (and the resulting expanding infrastructure) while keeping staffing and budgets flat. CIOs that are looking not only to run the business, but also grow and transform the business can start with fundamentally changing how they purchase and consume infrastructure.
Over the last decade, server virtualization has helped to significantly improve the compute utilization and operational models. Converged infrastructure builds on this trend by creating solutions that have a well-understood and balanced architecture including compute, storage, and networking components. Today the typical enterprise data center is a heterogeneous mix of solutions optimized to specific applications. Virtualization lets IT standardize on a solution that can support multiple applications with the virtual machine as the building block.
Of course resource requirements vary greatly depending on application requirements, so improving utilization even in a converged infrastructure environment requires an understanding of usage patterns. While virtualization is a major driver of this trend, it should be noted that few organizations are 100% virtualized today; so consideration should be given to whether a solution only supports virtualization or also bare metal environments.
More than simply just enabling virtualization, convergence solutions must support adoption of new operational models. It is not enough that converged infrastructure simply help IT improve the speed and agility of deploying infrastructure. For the overall economics of IT to improve, the automation and orchestration of convergence solutions must finally help move the needle from spending 70% of budget on “keeping the lights on” to less than 50%.
Reference Architectures and Integrated Solutions
Building solutions that include compute, network and storage are not new. The big IT players like HP and solution architectures such as IBM Redbooks have delivered bundles for many years. Convergence provides a spectrum of integration from less-integrated systems that allow customers or VAR/SI to choose piece parts while providing certifications or supporting matrix documents to guide the deployment. On the other end of the spectrum is a “Single SKU” environment where a solution is designed, tested, built, ordered, shipped, and installed as a single appliance. Oracle Exadata and Exalogic products are fully integrated offerings that are optimized for Oracle databases – a full stack of Oracle hardware and software along with the requisite licenses.
There are now many gradations along the spectrum from roll your own to single SKU. Reference architectures are the evolution of white papers and other best-practice publications. The general attacks on the various approaches are that fully integrated (single SKU) solutions are inflexible, while reference architectures cannot deliver on greater efficiency of infrastructure or improvements in operations. If we deconstruct the convergence options, it all comes down to where value is being added into the system and ultimately how it delivers on the promised goals of increasing overall OpEx.
EMC’s history with its convergence offerings provides good insight into the various options. As I wrote in The VCE Company, the original Vblock was a reference architecture designed and tested by EMC (along with partners Cisco and VMware) and built and shipped by channel partners. In order to reduce variability of configurations and increase the speed of deployments, VCE brought the building of Vblocks in-house; channel partners can perform software configurations prior to shipping. EMC is leveraging much of the knowledge and processes from VCE to create EMC VSPEX, a converged solution that the channel builds based on certified configurations.
Wikibon forecasts that by 2017, 2/3 of IT infrastructure will be delivered through converged offerings - totaling more than $250B in hardware, software, and services. In general, larger enterprises will have an affinity for higher integrated offerings, while the overall market for reference architectures remains larger.
Blade Servers and Beyond
In many ways, blade servers are the first generation of infrastructure convergence. Blade servers, especially when paired with server virtualization, allowed for simplified operations through a flexible architecture.
Simplified networking (and cabling) is an important part of the blade server value proposition. Cisco’s entrance into the compute market with its UCS platform was predicated on taking the compute and networking integration to the next level. Cisco’s partnerships with NetApp (Flexpod) and EMC (Vblock through VCE) have put it in a leadership position for convergence solutions.
IBM’s recent announcement highlights that convergence solutions will be disruptive to the blade server market.
Storage Angle of Convergence
Storage is the stickiest of the infrastructure components. While customers may not choose a convergence stack because of a storage preference, existing vendor installed base can slow or deter the adoption of a full stack solution if there is not a storage match.
Over the last 25 years, storage has moved away from compute into centralized, shared arrays. With the recent revolution of flash-based storage, we see that for certain workloads that is reversing, moving Tier 0 storage closer to the application to reduce latency and increase throughput. Flash storage is becoming integrated into compute architectures (such as the recently announced HP Gen8) and is starting to move into converged solutions. EMC is expected to announce a version of VFCache for Cisco UCS soon, allowing it to be integrated with VCE Vblock and EMC VSPEX solutions. Startup vendor Nutanix has flash fully integrated into its architecture, including both PCIe and SSD flash tiers.
In addition to flash, the other storage question for convergence is multi-vendor storage use. Oracle, Dell, EMC, NetApp, and HP all limit the storage platform that can be used in a convergence solution. Storage virtualization solutions from IBM and HDS are designed for heterogeneous storage allowing users to incorporate older competitive storage devices as a tier.
Shifts at the Edge of the Network
Cisco's entry into the compute market has stimulated the other compute vendors to offer non-Cisco options for integrated solutions. While convergence solutions do not have a significant impact on the core switching options that customers have, customer adoption of non-Cisco embedded switching is increasing. Customers should be careful that the vertical integration of stacks does not make the management of the entire network more complex. Converged solutions lean heavily towards Ethernet-based architectures inside the rack (Oracle being the lone InfiniBand-based solution), although FC connectivity is made available in a number of methods (blade for IBM PureSystems and at top of rack for VCE Vblock and HP VirtualSystems). Other than the potential shifts in vendors for embedded solutions, the impact of convergence on networking is not much different than for environments that are using virtualization and/or blade servers.
One Management to Rule them All
When choosing a converged solution, the orchestration layer, not the infrastructure components, has the greatest opportunity to increase an IT organization’s operational efficiency. These software tools allow the overall solution to go beyond being just a platform for virtualization infrastructure and allow enterprise IT users to act more like internal service providers. Most of the convergence solutions have management software that can orchestrate the entire stack. VMware vCenter and Microsoft SCVMM also provide varying levels of integration.
As an example, Dell’s vStart solution has two options for Private Cloud Management, the Virtual Integrated Systems portfolio from Dell and Microsoft’s System Center 2012. NetApp’s FlexPod leverages partner offerings for the orchestration layers through an open API so that customers can choose from BMC, CA, Cisco, Cloupia, DynamicOps, Gale, IBM, Microsoft, VMware, and others. Users should test drive the management tools and understand the ramp-up time each requires for the IT staff to master the new processes.
The Current Field of Solutions
|Oracle||Exadata and Exalogic||Single SKU = Oracle server and storage + InfiniBand networking from Mellanox||2011 Revenue: $1B|
|VCE||Vblock||Single SKU = Cisco UCS + Cisco Networking + EMC Storage + VMware||2011 Revenue: $209M, launched in 2009|
|NetApp||FlexPod (joint solution with Cisco)||Reference Architecture = Cisco UCS + Cisco Networking + NetApp Storage + VMware or Microsoft Hyper-V or Citrix||Launched in 2010|
|IBM||PureSystems family||Single SKU = Single chassis that has compute, networking and storage (primarily IBM with some options on networking) + broad choice on OS/hypervisor||Launched in 2012 (expected to ship in June)|
|HP||VirtualSystems||Single SKU = HP blade server + HP storage + HP networking + VMware or Microsoft Hyper-V||Launched in 2011|
|Dell||vStart||Converged offering = Dell rack server + Dell storage + Dell PowerConnect networking||Similar to Single SKU, single screen to order, Launched in 2011|
|HDS||Hitachi Unified Compute Platform||Single SKU = Hitachi Blade Server + Hitachi Storage||Launched in 2010|
|Nutanix||Nutanix||Single SKU = Converged Virtualization Appliance includes Fusion-io||Launched in 2011|
|EMC||VSPEX||Reference Architecture = EMC Storage + Intel-based servers + Cisco/Brocade networking + VMware or Microsoft Hyper-V||Launched in 2012|
Note that all vendors listed have numerous other, less-formal programs that fall into the “Reference Architecture” bucket.
Action Item: Converged infrastructure solutions are a different model of purchasing (will impact refresh cycles), managing and consuming IT. For customers who have moved aggressively along the journey towards virtualizing the data center, these solutions can have an immediate positive impact on speed and agility of scaling infrastructure. As with most new technology, a pilot deployment is recommended to learn the tool set and see how this new operating model fits with the IT staff. CIOs should look for technology and channel partners that can bring application expertise and a plan to improve operational efficiencies.
Infographic: Stack Wars: What is the Enterprise Stack
Stu Miniman is a Principal Research Contributor for Wikibon. He focuses on networking, virtualization, converged infrastructure and cloud technologies. Stu can be reached via email (firstname.lastname@example.org) or twitter (@stu).