CB308 – Real-Time Auditing & Defense
Lecture 5: Scheduling Periodic & Sporadic Tasks
Lecture 6: Multiprocessor Scheduling & Resource-Access Control
Lecture 5 — Scheduling Periodic & Sporadic Tasks
Slide 1 – Title & Context
• Scheduling Periodic & Sporadic Tasks
• CB308 – Real-Time Auditing & Defense
• Instructor Name · Date
Speaker Notes:
• Tip for students: Visualise periodic tasks like a
metronome; sporadic tasks are surprise claps between
beats.
Slide 2 – Learning Objectives
• Handle periodic & sporadic tasks reliably.
• Apply admission tests for sporadic requests.
• Guarantee deadlines under unpredictable arrivals.
Speaker Notes:
• Keep these three verbs in mind: HANDLE – TEST –
GUARANTEE.
Slide 3 – Arrival-Pattern Review
• Periodic exact inter-arrival T (e.g., 10 ms sensor tick).
• Sporadic ≥ minimum gap Tₘᵢₙ (e.g., push-button).
• Irregular arrivals demand bandwidth control.
Speaker Notes:
• Image placeholder →
Periodic_vs_Sporadic_Timeline.png (full-width).
Slide 4 – Periodic Task Parameters
• Each task: Period T, WCET C, Deadline D.
• CPU utilisation U = Σ C/T.
• RM is optimal among fixed-priority schemes for purely
periodic sets.
Speaker Notes:
• Mnemonic: ‘T-C-D’.
Slide 5 – RM Worked Example
• T1 (10 ms,1 ms), T2 (20 ms,2 ms), T3 (50 ms,5 ms).
• U=0.30 ≤ RM-bound(3)=0.78 schedulable.
• Shortest period → highest priority.
Speaker Notes:
• Image: RM_Gantt_Simple.png (centre).
Slide 6 – Sporadic Task Model
• Sporadic task has min gap T_s; uncontrolled arrivals risk
overload.
• Sporadic server gives budget C_s every T_s.
• Blends with RM or EDF scheduling.
Speaker Notes:
• Diagram: bucket analogy for bandwidth.
Slide 7 – Sporadic Servers in Action
• Capacity replenishes each T_s so backlog bounded.
• Better average response vs polling.
• Implemented in VxWorks POSIX sporadic server.
Speaker Notes:
• Show real trace if available.
Slide 8 – Admission / Acceptance Test
• Compute U_new = U_curr + C_s/T_s.
• Admit if U_new ≤ U_bound (RM) or ≤1 (EDF).
• Otherwise defer or reject request.
Speaker Notes:
• Image: Admission_Test.png (decision diamond).
Slide 9 – Case Study: Packet Filter
• Periodic routing refresh 50 ms (C=5 ms).
• Sporadic firewall checks (T_s=2 ms, C_s=0.3 ms).
• Utilisation kept <70 % to avoid stalls.
Speaker Notes:
• Image: Network_Sporadic_Case.png (centre).
Slide 10 – Key Takeaways & References
• Periodic = heartbeat; Sporadic = bounded bursts.
• Sporadic server + admission test protect deadlines.
• Read: Sprunt 1989; Buttazzo 2011; Liu 2000.
Speaker Notes:
• Quick quiz: What if T_s too small?
Lecture 6 — Multiprocessor Scheduling & Resource-Access Control
Slide 1 – Title & Architecture
• Multiprocessor Scheduling & Access Control
• CB308 – Real-Time Auditing & Defense
• Multi-core ≠ automatic speed-up!
Speaker Notes:
• Image: Multicore_Architecture.png (right).
Slide 2 – Learning Objectives
• Partitioned vs Global scheduling.
• Load-balancing heuristics.
• Priority Inheritance & Priority Ceiling protocols.
Slide 3 – Partitioned vs Global Overview
• Partitioned: offline bin-pack tasks; no migrations at
runtime.
• Global: single ready queue; tasks migrate.
• Trade-off: predictability vs utilisation.
Speaker Notes:
• Draw side-by-side diagrams.
Slide 4 – Partitioned Algorithms
• Heuristics: First-Fit, Worst-Fit, Best-Fit Decreasing.
• NP-hard → approximate solutions.
• Low migration cost; better cache locality.
Speaker Notes:
• Icon: bins and coloured blocks.
Slide 5 – Global EDF & Pfair
• EDF-Global: earliest deadline runs anywhere.
• Schedulable if U_total ≤ m – (m-1)·U_max.
• Pfair/LLREF: fairness by slicing quanta; high migration
cost.
Speaker Notes:
• Image: Global_EDF_Timeline.png (centre).
Slide 6 – Load-Balancing Strategies
• Work-stealing, CPU affinity, migration thresholds.
• Linux RT push/pull; Zephyr SMP idle-pull.
Slide 7 – Priority Inversion Problem
• Low-P task holds mutex, High-P task blocks, Medium-P
pre-empts.
• Mars Pathfinder resets every few sols.
Speaker Notes:
• Timeline sketch of inversion.
Slide 8 – PIP vs PCP
• PIP: mutex owner inherits highest waiter priority.
• PCP: global ceiling prevents deadlock & bounds
blocking.
• Blocking ≤ one lower-priority critical section.
Speaker Notes:
• Image: ResourceSharing_Protocol.png (full-width).
Slide 9 – Pathfinder Case Study
• Original kernel lacked inheritance ➔ reset loop.
• Patch added priority inheritance; fault resolved.
• Lesson: always compute worst-case blocking.
Speaker Notes:
• Image: Pathfinder_Fix.png (right).
Slide 10 – Takeaways & Sources
• Partitioned simpler; Global uses CPU better with
overhead.
• Proper resource protocols avoid deadly inversions.
• Read: Rajkumar 1990; Baruah 1999; Liu 2000.
Speaker Notes:
• Challenge: schedule 90 % load on 2 cores.
