Cloud computing uses externally maintained, commonly virtual infrastructure to provide computing and storage. Serverless arrangements can hide servers entirely. section 1.2.3
For performance-sensitive Python workloads, cloud is an execution and cost context, not simply a way to add machines or network storage. Proprietary resource extensions and APIs affect performance, virtual machines can make CPU-level tuning less predictable, and elapsed time directly affects cost. section 1.2.3
Account for these system-level implications:
- Virtualization can make cache-locality optimizations uncertain because another virtual machine may preempt the cache. section 1.2.3
- Proprietary cloud compute and storage solutions have specific APIs and behaviors, so their performance consequences may need separate consideration. section 1.2.3
- Traditional cluster concerns still apply, but some cloud-specific problems require separate treatment. section 1.2.3
Figure 1.4 Underlying hardware architectures must be taken into account when choosing high-performance coding solutions.
Relate cloud architecture choices to the broader principle that underlying hardware shapes high-performance coding decisions. section 1.4
