Companies today are spreading their applications across multiple clouds in a hybrid fashion. According to a recent IDC CloudView study among 6,000 IT and line-of-business executives whose organizations have adopted cloud technologies, 73% are implementing a hybrid strategy, which most defined as utilizing more than one public cloud in addition to dedicated assets.
What is the best way to prepare for a hybrid cloud strategy that touches multiple public and private cloud vendors?
There are many factors to consider, including these:
- How to choose the clouds in the first place
- Which mechanisms can be used to not only discover existing assets on each cloud but to construct some form of governance around them
- Which applications should run on each cloud
- How to migrate them there
- What an ongoing series of assessments over time looks like
Let’s consider each of the facets of preparing for the challenge of implementing a hybrid cloud strategy to maximize the value that this disruptive technology provides.
Choosing Clouds
In some ways, the first step in this process is the easiest. Most hybrid cloud strategies begin with one private and two public clouds. On the private side, VMware’s giant install base is the typical choice, although some form of OpenStack or AzureStack offers viable alternatives or additions.
On the public side, Amazon Web Services (AWS) has long been the leader in this space, but Microsoft has done an impressive job at gaining ground with its Azure product.
By default, then, a hybrid cloud strategy would likely involve VMware, AWS, and Azure, but there are nuances that might lead an organization to choose differently or take an approach that encompasses more than just these three. On the public side, for example, a corporate agreement with IBM or Oracle may lead an organization toward a different conclusion. Or, if geography will be important—either due to concerns about latency or country-specific data privacy regulations—a choice like Alibaba Cloud might be included.
Similarly, there are instances in which a company may want a low-cost private cloud based on OpenStack for development and testing workloads to supplement production hosting on VMware. Or, in an attempt to leverage knowledge and training for a strategy that already includes Azure, an investment in Azure Stack for the private cloud may make sense.
Tooling Help: Cloud Management Platforms
With clouds selected—each with their own unique APIs, command-line interfaces, and consoles—the next step in the journey is to find a cloud management platform (CMP). A CMP will provide tooling across this now-diverse set of clouds so that there is a single management control and governance point.
Critical features and value that a CMP should provide are embedded in each of the latter stages described below.
Existing Application and VM Inventory
Rarely does a hybrid cloud implementation start off with a green field absent of existing applications and cloud accounts. Creating an inventory of what the current applications are, where they currently run, and who is responsible for each of them is an important next step in the process—and also the first place a CMP can add value.
To make this discovery process go more smoothly, a good CMP can simply be pointed to an existing cloud account and be able to automatically discover the inventory of virtual machines (VMs) running there. Additional analytics tools can determine the nature of those VMs in a larger application context. For example, if one VM has traffic over ports 80 and 443, the tools can send some traffic to a second VM over port 3306 to determine that the first VM is likely a web server and the second is its database server.
Governance
Once the inventory of VMs and applications has been determined, the CMP selected should be ?able to synchronize with whatever single sign-on mechanism is already in use in an organization and obtain group definitions from an existing LDAP server. Combining these three sources allows an administrator to specify consumption limits for an individual or a group and assign the discovered brownfield VMs and applications to those individuals or groups.
For new deployments performed within the context of the CMP, the administrator can establish other limits that dictate who is allowed to deploy what applications where. It is critical to any hybrid cloud rollout to give constituents the ability to deploy resources in a self-serve, on-demand fashion—but in a way that gives an administrator some levers of control. This is exactly what the CMP provides from a single-control point across the several different cloud back ends.
Benchmarking What Goes Where
After creating a known brownfield inventory and a mechanism that dictates who can deploy what and where, it is necessary to determine the best place to deploy each application.
Aspects such as geography, data sensitivity, data gravity, and security all factor into that. The mistake that many people make is assuming that all clouds are created equal for all applications and that price is the exclusive metric that should be analyzed. Nothing could be further from the truth. Each cloud uses different hypervisors, networking topologies, disk arrays, and a multitude of other factors.
As such, it is unreasonable to expect, say, a .NET-based application to perform the same on AWS as it does on Azure. Microsoft has a home-field advantage with such applications, and the expectation would be that they would run better there. Other types of applications would have different answers across all the different cloud possibilities. If the price on one cloud is half that of another, but the performance is one-quarter as good, focusing exclusively on price would lead to a bad decision.
So, the only way to determine the price/performance metric is to benchmark an application on multiple clouds, and often on multiple instance types within a cloud, to find the right fit.
Migration
Once a destination has been selected for a particular application, the CMP should not only assist in deploying the application components on the new cloud but also provide hooks to migrate the data from cloud A to cloud B. Don’t think of this in terms of taking operating system image snapshots, as those are not typically compatible across multiple clouds and can get large in size. Instead, a more flexible approach is to think of the individual software components—such as web servers, database servers, and queuing servers—and use CMP automation or an existing investment from tools such as Chef, Puppet, or Ansible. Those are much easier to deploy and allow the application data to be dealt with separately.
Ongoing Assessments
It is important to realize that the cloud market is constantly changing. Even when benchmarking assessments are performed on an entire application portfolio, it does not mean that the right answer today will still be the correct one three quarters from now. Public cloud providers routinely introduce new instance types and open new regions. Private clouds frequently get new hardware deployed within them. Business needs shift over time, and while speed might be the most important factor at one point, cost may trump it later.
Generally speaking, every application should be reassessed annually, with more critical or costly applications evaluated more frequently. Additionally, news in the cloud space should be monitored weekly to see if something, such as a new instance type, needs to be benchmarked against the winner of the last scheduled assessment.
What Happens Next
Most organizations are headed toward a hybrid cloud strategy. Picking the clouds themselves is the easy part. Leading players such as VMware, AWS, and Microsoft provide a commonly used baseline, and factors such as geography, security, and corporate relationships lead some organizations to choose differently or have a greater number of choices.
The difficulty lies in what happens after the clouds have been selected, and this can be eased by a good CMP. Determining which applications and VMs already exist on different clouds is essential so that an organization can get a handle on just how big a problem it has. In addition, the ability to put limits on users without making them wait weeks through a byzantine ticketing system for access to resources is critical for a governing body to avoid a mutiny among those it is trying to serve.
When determining which applications run best where, do not fall into the trap of looking at price exclusively. It is critical to consider performance through benchmarking as well as factors such as data sensitivity, geography, and others to expose hidden costs. And, do not assume that the selection made today will remain the best choice for the long term as the market matures.
In the end, each hybrid cloud implementation is unique to the needs of the organization implementing it. These common phases involving cloud selection, brownfield inventory, governance, placement, migration, and ongoing assessment are helpful as a guide, but each organization will proceed on a slightly different path. That choice is what a hybrid cloud is all about.