For my last entry in this series on hyper-converged (HC) economics, I want to cover 2 points.
1. Where is the sweet spot for HC
2. Economic case study comparing HC to CI
So first, where is the sweet spot for appliance based HC systems?
- From a practical point of view, we see a lot of appliance HC systems used for VDI deployments. The growth here tends to be linear, and the workload well defined. As more VDI instances are added, the building block appliances can easily be purchased and installed.
- Generally, lots of small VM systems may also gravitate to HC appliances. When a client has mostly small (1 or 2 vCPU with 1-2GB of memory each), a good case can be made for HC appliances. The key is the size and growth rate of the storage per VM. The following graphs come from running a VMware script against the current VM and host environment to see the sizes and ranges of VM assets within a particular infrastructure.
- In this fist example below, the VM size (lots of small systems) and lower GB of storage per VM, would economically favor HC appliances. There are lots of x-small and small VM, each with very low storage requirements
- This this example, even though there are lots of small systems, the storage capacity and growth rates would eventually create waste issued (see previous blog on non-correlated growth) in the appliance approach. Storage scale-out tends to cause waste in appliances.
- Finally, in this example the size of VM are such that too many appliances would be required to facilitate the large and x-large VM systems, and the rate of waste would be prohibitively high. Also with the large storage pools per VM, a traditional CI system with scale-out storage would be best for the overall architecture.
- VM with a short-term or very limited data protection requirement
- VM environment where organic growth is somewhat predictable, and the size of VMs and storage is also organic/predictable
Economic case study comparing CI to appliance based HC
We had a large US banking customer that had installed a CI (UCP system) a few years back, and were very happy with the management, support, performance and availability. Like all customers, a new lower-cost option with an applianced-based solution seemed compelling. The purchase price was lower than the UCP for similarly-size domains (each with 128 blades supporting 1.6 PB of storage and 1,400 VM of various sizes). See below for the VM size distribution
The economic analysis was to look at both purchase price and the total cost per VM over a 3 year period. The appliance solution had an initial lower cost per blade, and per CPU. The UCP was 3% higher in total cost of acquisition, and 7% higher on the cost per vCPU rate. As we looked beyond the purchase price, into the total costs of the VM, we were able to document a different TCO story. In packing all the VM into the bank’s system, the UCP solution delivers a 21% lower cost per (average) VM. One of the key factors was the footprint of the environment. The appliance approach required 18 racks in the data center, where UCP required only 8. The difference in floor space, power and cooling was $600K over the term of the assessment. Furthermore, up-front (one-time) engineering, certification and multi-year management labor added to the labor required for the appliances.
The cost elements that were included in the above multi-year cost per VM comparison were:
- HW and SW maintenance
- Storage and VM operational management
- Power and cooling, floor space
- Engineering and certification, patch mangement
- Cost of waste, cost of scale-out
- Networking costs (local and long-distant)
- VM license fees, OS, application software
- Cost of performance
- Cost of provisioning time
- Outage risk
- Data protection, disaster protection
- Data archive, long-term retention for the bank
In looking at the economics of different solutions for VM deployment (CI, cloud, HC, Do-it-yourself), we have to look beyond the purchase price to understand the total cost of these architectural decisions. Lower initial price can be seductive, but may lead to longer-term costs of ownership. Like any new architecture, HC has unique qualities and capabilities and its own economic sweet spot and sour spot. Be sure to consider the size, growth and scale-out requirements, as well as data protection and DR protection in considering the right architecture for your workload.