During a hybrid cloud proof of concept, it's important to test data migration processes, operational procedures...
-- such as failover -- and the integration points between private and public clouds. This ensures IT teams catch any issues around latency and error handling on the data side, as well as around cloud bursting and load balancing on the operational side.
Here are four steps to move forward with a hybrid cloud proof of concept (POC) in your enterprise.
1. Evaluate hardware changes
First, assess the impact of hardware changes. Most likely, your hybrid cloud will involve the migration of an existing VM server cluster to the public cloud, a new in-house setup or both. In any case, the initial migration will generally be from older gear to state-of-the-art servers. While virtualization abstracts away the hardware to a great extent, this will introduce some change. For example, compute instances will have memory differences. Additionally, available solid-state drive space and performance -- together with LAN bandwidth -- will likely increase.
Public cloud platforms evolve constantly, as reflected by providers' ever-expanding menu of instance types. Use a sandbox environment to optimize instances, as well as costs. This also lets you right-size and balance your initial hybrid cloud deployment.
2. Data migration
The next step toward a hybrid cloud POC is data migration. Purge any unnecessary files, and map out which data sets will live where in your deployment. Some will only be in-house, some will only be in the public cloud and the rest will cross the cloud boundary. Use governance and compliance software to validate these decisions and save time.
The next step is the actual migration, which IT teams generally perform in one of two ways:
- Piecemeal method: Also referred to as the trickle method, this approach occurs in phases and often without shutdowns. This does, however, require applications to synchronize all the data and is more difficult from a testing perspective. IT teams that choose this migration approach might need to perform additional tests in parallel operation.
- Big bang approach: This option requires a day or two of downtime, as IT repositions all the data. Compared to the piecemeal approach, this migration method is less technically complex but can feel much more like a race against the clock before restart time.
Operations teams should drive app migrations, as well as the migration of the operating environment, including network structures and storage. In most cases, apps will run in the cloud the same way they ran in the legacy environment, but certain changes, such as a move from block to object storage memes or even a general version upgrade, could make things complicated. Don't make any changes to an application during the migration process, as this could create issues around debugging or even lock that application into the new public cloud platform.
3. Set up networks and cloud bursting
Another challenge IT teams will face leading up to a hybrid cloud POC is building scripts to set up virtual networks in the public cloud. While certain hybrid cloud platforms, like Azure Stack, make this more seamless, a general lack of cross-cloud standards complicates this process. There are, however, translation tools for scripts, such as OpenStack Heat and AWS CloudFormation, and the industry is moving closer to interoperability standards.
With hybrid cloud networks, cloud bursting is often a top priority for enterprises, which means it should be a day-zero feature -- one that's available to use immediately. But before you commit a bursting process to a hybrid cloud POC, test it extensively in a sandbox. There are a variety of models for cloud bursting, some of which involve dedicated public cloud instances and some that use on-demand instances. Develop a careful financial model to ensure you fully optimize cost during bursts to the public cloud, and lean toward doing that after day zero.
4. Resiliency testing
Finally, once the hybrid cloud POC appears to run properly, you need to test its resilience and failover processes. Resilience refers to the ability of the compute cluster to lose nodes but still continue operations, as mediated by your orchestration system. Failover, on the other hand, occurs in more catastrophic circumstances, such as a cloud's availability zone going down for several hours.
For a successful hybrid cloud deployment, it's critical to test both of these processes. Success ultimately depends on the resiliency standards you set and will influence how you position snapshots and replicate data sets.
At this phase of migration, IT budgets and schedules are usually under pressure, but try not to cave in. It's important to test failover multiple times.