In a previous article, I suggested that hyperconverged is just a step on a path to simpler IT infrastructure. Before I look at the specific technology areas that I think will become simplified, I’d like to look at how the simplification works. I see simplification coming in two parts. The first is very advanced complexity, in which there is a lot of automated complexity that removes human decision-making. The second is making the required human activities as simple as possible.
VMware’s Distributed Resource Scheduler (DRS) is an example of advanced complexity. DRS moves VMs between hosts in a cluster as required for the VMs to get the resources they are entitled to receive. It takes a lot of complexity to achieve DRS functionality, but there is little complexity to operating DRS. A basic DRS configuration requires about three settings:
- DRS – Enabled
- Automation Level – Fully Automated
- Migration Threshold – Default
These settings are essentially a basic resource-management policy for the cluster. Behind the covers, DRS uses these settings and evaluates the cluster on a regular basis. DRS automatically takes any required actions. The operations team does not need to monitor through the day to determine whether DRS is doing its job. There are a heap of prerequisites for DRS to work. That seems an ideal next place for simplification. But once DRS is operational, the operations teams can spend less time managing individual hypervisor loads. They can focus more at the cluster level. This elevation out of the minutiae and broader infrastructure focus is a crucial part of simplification. All of the HCI vendors are building a great deal of complexity into their platforms, but they are hiding it from users with simple interfaces.
Much of simplifying user activities comes down to user interface (UI) design and consolidation. Having everything consolidated into a single UI simplifies routine tasks. It isn’t realistic to expect a single pane of glass for an enterprise organization: there will be many single panes of glass. Enterprise organizations will also want the ability to automate. Infrastructure APIs and SDKs are crucial. For a small business, the infrastructure could be managed from a small number of tools: hopefully, one primary interface that integrates all routine tasks, and a small number of less frequently used tools. Any single pane of glass needs to streamline routine tasks, removing unnecessary steps and options. The UI needs to be tightly focused on completing tasks so the administrator can move on to the next task.
We see simplification driving architectural decisions in hyperconverged. One example is converging storage networking into the Ethernet network. By eliminating a dedicated storage network, the required skill sets can be reduced. Traffic segmentation, security, and performance are managed quite differently on Fibre Channel to Ethernet. Having one network type simplifies network management.
To be clear, the requirement to manage does not go away. However, one set of domain-specific knowledge (usually Fibre Channel) is removed. We also see this simplification in the choice of NFS or another proprietary format over iSCSI for shared storage. This removes the need to manage LUNs and VMFS file systems along with LUN and device queue depths and multipathing. The software-defined storage (SDS) component also removes the need to manage RAID, tiering, and caching. The advanced complexity in the platform manages all of the performance aspects. Again, a technology decision is made to simplify ongoing operations activities.
Policy-based management is one of my essential capability areas for simplified IT. There should be no need to continually check that things are working as they should. The operations team should only need to investigate noncompliance. We see this paradigm in declarative state tools like Puppet and Chef. The policy identifies the desired state of the infrastructure. The infrastructure then ensures that it is policy compliant or raises alerts if it is unable to achieve compliance. The policy should define the required state; manual state changes should be unnecessary. When a change is needed, the policy should be updated, so the change is applied automatically.
There are many areas of IT infrastructure simplification that are yet to be addressed by hyperconverged vendors. Over time, we will see more features and functionality integrated into these platforms. We will see less need to spend time managing infrastructure and more focus on the applications and data that deliver business value.