Which of the following describes major OS-level performance bottlenecks and practical mitigation strategies such as batching I/O, prefetching, lazy loading, concurrency, and caching?

• A. The main bottlenecks are GPU cycles and network latency; mitigations include increasing GPU cores.
• B. The main bottlenecks are CPU cycles, memory bandwidth, and I/O latency; mitigations include caching, prefetching, I/O batching, asynchronous I/O, memory management tuning, and parallelization.
• C. Bottlenecks are unpredictable; there are no general mitigation strategies.
• D. The main bottleneck is unit test time; mitigation is test parallelization.

Answer: (B)

Performance at the OS level is most affected by how quickly the system can compute, move data, and wait for devices. The primary bottlenecks are CPU cycles, memory bandwidth and latency, and I/O latency. When CPU time is the limiter, work spends too long in processing and scheduling; when memory bandwidth or latency is the limiter, moving data between caches, RAM, and memory controllers becomes the hold-up; when I/O latency dominates, waiting on disks, networks, or other devices slows everything down.

Mitigations that fit these bottlenecks target reducing waits and increasing overlap. Caching stores hot data in faster memory so repeated access doesn’t hit the slower storage again. Prefetching and lazy loading anticipate needs and defer or gather data just in time, hiding latency. I/O batching reduces overhead per request and improves throughput, while asynchronous I/O allows the program to continue work while I/O completes. Memory-management tuning optimizes how the system uses page caches, swapping behavior, and allocator strategies to improve data movement. Parallelization spreads work across multiple cores and I/O queues, enabling concurrent processing and overlapping of computation with I/O.

These strategies align with the typical OS-facing limits. Other scenarios might involve different constraints, but CPU, memory, and I/O remain the central OS-level performance levers for most workloads.