Cloud services provide a variety of computing options but getting the best value should start with initial compute requirements.
CPU Performance Considerations
- Chose processor models that are optimized for virtualization (Intel-VT or AMD-V).
- Determine the number of cores based on the number of app and the needs of each app to run on the host.
- Heavily used apps will require more cores than non processor-intensive applications.
- In public and hybrid cloud deployments, right-sizing CPUs, rather than paying for more cores, will keep costs and energy use down.
- Consider staying with one brand of CPU (Intel vs AMD) for easier VM migrations and management.
CPU core increases do not scale as CPU clock-speed increases. A 3.2 GHz CPU is twice as fast as a 1.6 GHz CPU, a quad-core CPU is notfour times faster than a single core. In terms of speed, dual core CPU is roughly 50% faster than a single-core, and a quad-core CPU is only about 25% faster than a dual-core.
VMWare has guidelines to commit ratio calculations:
- 1:1 to 3:1 is not a problem
- 3:1 to 5:1 will begin to cause performance degradation
- 6:1 or more will cause issues
CPU Sizing Considerations in Cloud Deployments
- Moving VMs from one processor architecture to another will cause issues. Moving from an AMD to Intel will cause the VM to crash, so it’s a good idea to stick with a single AMD or Intel architecture.
- Consider dedicated hosting for high I/O apps (big data).
- Peak usage should guide the decision when assigning vCPUs.
- Always factor in future growth as well as current usage.
Memory Performance in Cloud Deployments
When allocating memory for VMs, consider only giving the exact amount of memory it needs at peak loads. Increasing the memory in a cloud environment is a simple task and by only allocating the exact memory the VM needs, reduced costs will be achieved. Continue to monitor memory allocation to eliminate over-committing or under-committing memory usage.
Similar to CPU energy efficiency savings, consider energy usage when sizing memory.