CSI Performance Benchmarking & Tuning 3 — Questions and Answers
Question 1: What does the acronym 'IOPS' measure in storage performance benchmarking?
- Input/Output operations per second (Correct answer)
- Internal operating processor speed
- Indexed object partition size
- Integrated output packet sequence
Correct answer: Input/Output operations per second
IOPS measures how many read or write operations a storage device can perform per second, a key metric for evaluating storage performance.
Question 2: During a stress test, which condition signals that a system has reached its 'breaking point'?
- Error rate rises sharply and throughput drops simultaneously (Correct answer)
- Response time increases by more than 10%
- CPU utilization exceeds 50%
- Memory usage grows by 20%
Correct answer: Error rate rises sharply and throughput drops simultaneously
The breaking point is characterized by both a spike in errors and a drop in throughput, indicating the system can no longer handle additional load.
Question 3: Which HTTP keep-alive setting tuning reduces connection overhead in a high-request-rate web application?
- Increasing the keep-alive timeout and maximum requests per connection (Correct answer)
- Disabling keep-alive to force fresh connections
- Reducing the TCP window size
- Setting connection timeout to zero
Correct answer: Increasing the keep-alive timeout and maximum requests per connection
Increasing keep-alive timeout and max requests per connection allows more requests to reuse established TCP connections, reducing handshake overhead.
Question 4: A system integrator notices that adding more threads beyond a certain count actually decreases throughput. Which principle explains this?
- Amdahl's Law and thread contention overhead (Correct answer)
- Moore's Law limits on transistor density
- CAP theorem trade-offs
- Little's Law queue balancing
Correct answer: Amdahl's Law and thread contention overhead
Amdahl's Law predicts diminishing returns from parallelism, and excessive threads introduce contention overhead that outweighs the parallelism benefit.
Question 5: What is the main advantage of using percentile-based SLAs (e.g., p99) over average-based SLAs for latency targets?
- Percentiles capture tail latency experienced by the worst-affected users (Correct answer)
- Percentiles are easier to calculate than averages
- Percentiles always produce lower numeric targets
- Percentiles eliminate the need for load testing
Correct answer: Percentiles capture tail latency experienced by the worst-affected users
Average latency can mask severe slowdowns for a subset of users, while percentile-based SLAs ensure that even the worst-case users receive acceptable service.
Question 6: In network performance tuning, increasing the TCP receive window size primarily improves:
- Throughput on high-latency, high-bandwidth links (Correct answer)
- DNS resolution speed
- Packet loss recovery time
- Firewall rule processing speed
Correct answer: Throughput on high-latency, high-bandwidth links
A larger TCP receive window allows more data to be in flight before an acknowledgment is required, which maximizes throughput on high-latency links.
Question 7: Which of the following best describes 'think time' in a load testing scenario?
- Simulated delay between user actions to mimic real user behavior (Correct answer)
- Server processing time for each request
- Network round-trip latency
- Time required to parse a server response
Correct answer: Simulated delay between user actions to mimic real user behavior
Think time is an intentional pause inserted between virtual user actions to simulate the time a real user spends reading or processing a page before acting again.
CSI Performance Benchmarking & Tuning 3
What does the acronym 'IOPS' measure in storage performance benchmarking?