Lecture 8a

Question 1

Symmetric Multiprocessor Issues

Answer 1

Tradeoff between load balancing and processor affinity

Question 2

Processor Affinity

Answer 2

• When a process runs on a processor, some data is brought into that processor’s cache
• Process migrates to another processor
• Cache of new processor has to be repopulated
• Cache of old processor has to be invalidated

Question 3

Load Balancing

Answer 3

• Ensure load is evenly distributed among processors
• Balance load using push migration (a specific task periodically checks load on processors and evenly distributes it evenly by moving tasks) and pull migration (idle processor pulls a waiting task from a busy processor)

Question 4

Hard Real-time Scheduling

Answer 4

• Task must be finished within a deadline

- Need different types of schedulers to ensure that deadlines are met (earliest deadline first)

Question 5

Soft Real-time Systems

Answer 5

• No strict deadline, but should be executed “quickly”

- Priority based on scheduling with preemption

Question 6

Linux: Completely Fair Scheduler (CFS)

Answer 6

• For scheduling lower priority processes
• Select task with lower virtual run time
• O(log N) with red-black tree, and O(1) with caching

Question 7

CFS Target Latency

Answer 7

• Time during which every runnable task should run at least once
• Each process gets a time portion based on its nice value (from -20 to 19, lower nice = higher priority)

Question 8

CFS Virtual Run Time

Answer 8

• Used to automatically determine priority
• Physical run time + decay (lower priority processes have higher decay)
• Priority inversely proportional to virtual run time

Question 9

Deterministic Modeling

Answer 9

• Run all algorithms on a workload

- Not general

Question 10

Queuing Models

Answer 10

• Use queuing theory to analyze algorithms

- Many (unrealistic) assumptions to facilitate analysis

Question 11

Simulation

Answer 11

• Build a simulator that models systems
• Use a synthetic workload or traces from real systems
• Expensive (can take hours/days)

Question 12

Implementation Evaluation

Answer 12

• Code an algorithm and test it

Question 13

Why do processes cooperate?

Answer 13

• Information Sharing
• Computation Speed-up
• Modularity, convenience

Question 14

Interprocess Communication (IPC) Methods

Answer 14

• Shared Memory

- Message Passing

Question 15

Shared Memory

Answer 15

• One process creates shared memory
• Other processes attach shared memory to their own address space
• Shared memory is treated as regular memory
• Synchronization is needed to prevent conflicts

Question 16

Shared Memory Pro/Con

Answer 16

• Pro: Fast (memory speed), convenient to programmers (just regular memory)
• Con: Need to manage conflicts (tricky for distributed)

Question 17

Message Passing

Answer 17

• Process A sends message to B via kernel
• send(msg), receive(msg)
• Direct vs Indirect (naming vs ports, mailboxes)
• Blocking vs non-blocking
• Buffering

Question 18

Message Passing Pro/Con

Answer 18

• Pro: No conflict (easy to exchange messages, especially in distributed systems)
• Con: High overhead and slow