February 24, 2016
This post is part of a series based on the forthcoming book, Building a Modern Data Center, written by Scott D. Lowe, David M. Davis and James Green of ActualTech Media in partnership with Atlantis Computing.
Hyperconvergence is a hot button topic in many IT organizations today. Much the same as software-defined storage (which we discussed in a previous article), there’s a bit of confusion in the industry around what constitutes a ‘hyperconverged’ architecture. Also, as this model is new to many data center consumers, there are some potential pitfalls that one should be made aware of when considering options for hyperconverged infrastructure (HCI). Let’s start by clearly defining hyperconvergence, and once that foundation is established, we’ll look at some of the considerations when choosing a hyperconverged infrastructure solution.
The term hyperconvergence has been thrown around a lot lately. Plus, there are other similar terms being used, including Server SAN, converged infrastructure, hyper-convergence, and more. So what really is hyperconvergence? The term ‘hyperconvergence’ breaks down into hyper (meaning quickly or with a hypervisor) and convergence (meaning bring together). Therefore, you can loosely translate hyperconvergence to mean “quickly bring together with a hypervisor, or software.” In the case of today’s hyperconvergence solutions, they bring the storage into the compute, although the hyperconvergence of the future may include the network, or more.
The servers run a hypervisor, and that hypervisor either runs the virtual storage appliance (VSA) or uses a hypervisor-integrated storage management module (which we discussed in the previous article) in order to provide the distributed storage layer for the virtual machines. That distributed storage layer is spread across the servers that make up the storage and compute cluster.
Distributing the storage across the compute layer using software-defined storage (SDS) and running your business workloads alongside the VSA is the minimum requirement to dub something “hyperconvergence.” However, most modern hyperconvergence solutions also provide:
“Single Pane of Glass” Management — A single, integrated management interface for both the virtualization and storage resources, all in one place, for administration, performance/capacity monitoring, and troubleshooting.
Simplified Packaging and Support — A single vendor from which to purchase the hardware and software (including virtualization and storage), as well as a single support contract to a single support group that will support the entire hyperconverged infrastructure (which included server hardware, software-defined storage layer, and virtualization layer).
Advanced Data Services — For example, deduplication, compression, I/ O acceleration, snapshots, cloning, and replication. Many hyperconvergence solutions provide these types of advanced features all in the software along with the solution.
SDS in Hyperconvergence
Increasingly, hyperconvergence is being driven with greater and greater innovations in the area of software-based advanced data services. Think of this as, “What else can the software-defined storage do, other than just distribute the storage across the compute?” Just as server virtualization provided numerous management efficiencies, there are numerous ways that SDS could make the lives of data center admins easier.
For example, when learning about and shopping for SDS solutions, you should ask how can the SDS solution:
Reduce data being stored through intelligent deduplication and compression, saving storage costs?
Accelerate I/O to allow you to virtualize applications that weren’t candidates before (due to their high I/O demands), and/or allow you to consolidate more virtual machines on the same compute/storage cluster in order to provide a greater return on investment?
Provide integrated data protection through offsite replication and cloud backup options?
Eliminate existing third-party solutions to provide you greater management efficiency and greater return on investment?
The Hyperconvergence Design Model
In many cases, too much is made out of how the storage layer is designed — either with VSA or hypervisor-integrated storage. In the end, neither the applications nor administrators will think about this design choice on a daily basis. What matters is the efficiency of the hyper-converged system in terms of how many compute and storage resources are available to workloads. To be candid, there are more important hyperconvergence implementation options such as…
Hybrid or All-Flash
Of particular importance when choosing a hyperconvergence solution are the types of storage media and how they are used in the hyperconverged infrastructure nodes. Hyperconverged solutions can use all hard drives, hybrid (mix of flash and hard drives), all-flash, or DRAM with flash.
The number of hard disks versus flash disks is going to have a number of business impacts, because it determines:
The number of virtual machines that can be run on the hyperconverged infrastructure.
The performance that the hyperconverged infrastructure provides the applications and the ability of the solution to run highly intensive applications such as virtual desktop infrastructure (VDI) or database analysis.
The cost of the hyperconverged infrastructure solution — the more flash disks, the higher the cost of the solution in most cases. However, the more flash disks the nodes have, the greater the performance and the more virtual machines you’ll be able to run. Thus, having more flash disks (or all flash disks) may actually more than pay off in the end, because that additional storage I/O performance that you’ll achieve can be used to run additional virtual machines. This, in turn, provides you a greater return on investment from your hyperconvergence investment.
However, the cost comparison isn’t always that simple. Many modern hyperconvergence solutions utilize pre-process deduplication in order to significantly reduce the flash storage requirements and, thus, reduce the cost. In many cases, the cost of an all-flash hyperconverged infrastructure can be reduced to the point that it costs less than hyperconverged infrastructures that utilize hybrid storage.
Additionally, modern hyperconvergence solutions can also use memory to eliminate the pre-process deduplication overhead and even accelerate storage I/O performance for the the data stored in the flash infrastructure.
Appliance vs. Software/Reference Architecture
When choosing a hyperconvergence solution you’ll find that numerous hyperconverged infrastructure solutions are offered as integrated “single SKU” appliances (essentially a physical server with a disk, virtualization layer, and distributed storage layer on top). Or they are offered simply as software-only solutions that you can choose to implement yourself, on whatever hardware you choose.
For the hyperconverged infrastructure solutions that are offered as software-only, many of those offer a reference architecture. Reference architectures tell enterprise architects and admins that if they configure the hyperconvergence software with a specific configuration of CPU, memory, and storage, how many virtual machines or what type of performance the enterprise will receive from that configuration. Reference architectures might also be as specific as to state what you can achieve if you use a specific brand and model of server, with a specific configuration. Then, if the enterprise truly uses the architecture as specified and doesn’t receive the expected outcome, the hyperconvergence vendor will support and troubleshoot the performance of the hyperconverged infrastructure. In other words, you can think of reference architectures as blueprints for a hyperconverged infrastructure that are defined by the hyperconvergence vendor in cooperation with hardware server vendors.
Both of these approaches having their own sets of pros and cons. With the “single SKU” packaged hyperconverged infrastructure appliance approach, the enterprise doesn’t have to think about server hardware, compatibility, or even sizing. Additionally, with the appliance approach, if there is a problem in the infrastructure, they will have a single group to call under a single support contract. The downside with the packaged hyperconverged appliance approach is that they are essentially locked in to only specific, static, server configurations from their hyperconvergence vendor with few options for price negotiation.
With the software-only, reference architecture approach, enterprises likely have the option to use their existing servers or at least obtain servers from their existing vendor/VAR at previously negotiated discount levels. This maintains the enterprises’ comfort-level while gaining the best of both worlds: leveraging a pre-architected, pre-sized, and fully supported blueprint to build their own hyperconverged infrastructure, while using their own hardware.
Also with a software-only approach, reference architecture approach, enterprises have the flexibility to leverage whatever hardware that they choose and that gives them the option to use the latest and greatest, highest performance, densest server configurations possible. For example, with software-only, enterprises could run hyperconvergence on blade servers or on super-dense servers with massive amounts of memory and CPU. Also, with the software-only approach, enterprises have the option to either scale up or scale out by either adding more CPU and memory to existing hosts or by adding more hosts of varying sizes.
Another aspect of hyperconvergence selection is whether the solution offers a choice in hypervisor support. Any solution that is hypervisor-integrated is only going to be offered on the hypervisor that the integration was done on. It’s the virtual storage appliance (VSA) design of many hyperconvergence solutions that support multiple hypervisors and offer the most hypervisor choice. We should clarify that even hyperconvergence solutions that support multiple hypervisors typically don’t support a heterogeneous hypervisor cluster (a single cluster made up of hosts utilizing different hypervisors).
While the hypervisor that you are using today may be supported by the hyperconvergence solution you are considering, it’s important to remember support for multiple hypervisors in your selection criteria as these hyperconvergence solutions will give you the greatest leverage when it comes time to negotiate software maintenance/support on your existing hypervisor (whether you choose to exercise that choice or not).
Server Hardware Choice
As mentioned above with the comparison between “Appliance vs. Software/Reference Architecture,” when considering hyperconvergence solutions, it’s important to ensure that they also offer you hardware choice. You may have little choice in hyperconvergence solutions that are sold or packaged as an integrated appliance (some vendors offer you the choice between two brands of appliances), but will have ultimate flexibility with hyperconvergence solutions that are sold as software-only/reference architecture.
While some form of lock-in is inevitable when choosing hardware and software solutions, enterprises want to ensure that the solution that they choose offers them choice or the flexibility to easily move to the alternative (whether they ever choose to exercise that choice, or not). In other words, very rarely is any enterprise ever really “locked in” and unable to make a change in hardware or software but the real question is, “What is the price to change?” The level of lock-in is very different if the price to move is zero versus if the price to move is $10 million dollars.
When considering choice in the context of server hardware, however, there are many organizations that want someone to tell them what to do. It’s easier for these companies to simply buy a complete hardware and software bundle that they can put it a rack, turn on, and immediately begin using. While they may become locked in to that platform, that is by choice.
Modern Data Center Experiences
In the next section in this series, we’re going to be taking a look at some real work data and case studies that explore and expand on how what we’ve covered so far in this series is showing up in the real world. If you’ve thought that all of this sounds nice, but how it all plays out seems a little vague, the next article is for you. Don’t miss it!
SCOTT D. LOWE – Contributor
Scott is Co-Founder of ActualTech Media and serves as Senior Content Editor and Strategist. Scott is an enterprise IT veteran with close to twenty years experience in senior and CIO roles across multiple large organizations. A 2015 VMware vExpert Award winner, Scott is also a micro-analyst for Wikibon and an InformationWeek Analytics contributor.
Current rating: 5 (2 ratings)