List of Figures

• 2.1. State transitions for a conventional scheduler
• 2.2. A traditional work-conserving scheduling framework
• 3.1. No deceptive idleness
• 3.2. Deceptive idleness due to explicit request dependencies
• 3.3. Deceptive idleness due to shared metadata
• 3.4. Deceptive idleness due to memory contention
• 4.1. Deceptive Idleness on a seek reducing scheduler: (a) effective seek reduction, and (b) thrashing due to deceptive idleness
• 4.2. Deceptive Idleness on a proportional-share scheduler: (a) effective 1:2 allocation, and (b) skewed 1:1 allocation due to deceptive idleness
• 4.3. Proportional-share scheduler delivering desired proportions
• 4.4. Deceptive idleness due to explicit request dependencies
• 4.5. Deceptive idleness due to shared metadata
• 4.6. Deceptive idleness due to memory contention
• 4.7. Lottery scheduler behaving as desired
• 4.8. Lottery scheduler achieving skewed proportions due to deceptive idleness
• 5.1. Impact of asynchronous prefetch on seek-reducing scheduler performance
• 6.1. NWCS system architecture
• 6.2. State transition diagram for the NWCS waiting mechanism
• 7.1. Different access patterns
• 7.2. Increasing thinktimes for both processes
• 7.3. Increasing thinktimes for one process
• 7.4. Adversary application
• 7.5. Seek-reducing proportional-share scheduler
• 7.6. Andrew Benchmark
• 7.7. The Apache webserver
• 7.8. The GNU Linker
• 7.9. The TPC-B database benchmark