Soft RealTime Scheduling on Simultaneous Multithreaded Processors

1
Soft Real-Time Scheduling on Simultaneous
Multithreaded Processors

• Rohit Jain,
• Christonpher J. Hughes
• Sarita V. Adve
• IEEE REAL-TIME SYSTEMS SYMPOSIUM (RTSS02)
• Borrowed from Yen-Sheng Chang
• and presented by Cristiano Pereira
• 4/22/2005

2
Outline

• Introduction
• SMT review (hyperthreading)
• Resource Sharing Algorithms
• Co-Scheduling Algorithms
• Experimental Methodology
• Results
• Conclusions

3
Introduction

• Simultaneous multithreading (SMT) improves
  processor throughput by processing instructions
  from multiple threads each cycle.
• Two Decisions with SMT
• Co-schedule selection
• Affects threads utilization
• Resource sharing
• Which threads get share processor resources among
  co-scheduled threads
• Problem unique to SMT processors
• The choice of co-scheduling and resource sharing
  algorithm may be tightly coupled.

4
Introduction - Objective

• To find the best algorithm to increase
  schedulability of soft real time tasks by
  exploiting co-scheduling and resource sharing on
  SMT processors

5
SMT review
simultaneous multithreaded
traditional (single-issue)
superscalar
multithreaded
6
Related Work

• ICOUNT (seeks to maximize throughput)
• Without consideration of any real-time deadline.
• Symbiotic Job Scheduling for SMT
• One interactive task with other non-real-time
  tasks
• Multiprocessor scheduling (without SMT)

7
Resource Sharing Algorithms (1)

• Threads share most processor resources
• Instruction fetch mechanism
• Instruction window
• Functional units
• Caches
• Previous work has focused resource sharing
  algorithms to maximize total throughput (IPC)
• No deadline concerns
• May have negative/positive impact on the
  schedulability

8
Resource Sharing Algorithms (2)

• Throughput-driven (dynamic)
• ICOUNT, gives priority to the thread that has the
  least instructions in the instruction window
• Performance prediction is difficult.
• Performance guarantees (static)
• Fixed set of resources is reserved for a given
  job
• May be suboptimal!
• Performance prediction is easy (identical to
  uni-processor).
• Resources controlled per thread in this work
• Fetch bandwidth instruction window (ICOUNT)
• Other resources (thread blind, e.g. give FU to
  oldest inst.)

9
Co-scheduling algorithms (1)

• Partitioning (bin-packing flavor algorithms)
• Allows admission control
• Global scheduling
• Task migration on an SMT processor is free.
• Symbiosis-aware vs. Symbiosis-oblivious
• On SMT processors, execution time of a job
  depends on jobs co-scheduled with it.

10
Co-scheduling algorithms (2)

• Design space explored
• EDF as the underlying algorithm.
• Partitioning ? EDF schedules the tasks within a
  context
• Global ? EDF chooses the next task.

11
Predicting Execution Time, Utilization, and
Symbiosis (1)

• These algorithms need to know exec. time,
  utilizations and symbiosis relations

IPC instruction count
12
Predicting Execution Time, Utilization, and
Symbiosis (2)

• IPC
• Job IPC in single-thread mode
• ? profiling one frame of each frame type.
• Job IPC with static resource sharing
• ? profiling each allocation in single-threaded
  mode
• Job IPC with dynamic resource sharing
• ? profiles all possible co-schedules (N-tuples)
  to obtain the task IPCs.
• Average job IPC with dynamic resource sharing
• ? when the IPCs depend on the yet unknown
  co-schedule, approximate as job IPC averaged
  across all possible co-schedules
• Instruction count
• Use the average instruction count of a large
  number of frames as the prediction.

13
Partitioning Algorithms (1)

• A partition in SMT is a set of tasks such that no
  two will execute simultaneously.
• SMT supporting with N contexts has up to N
  partitions

14
Partitioning Algorithms (2)

• PART-NOSYM-DYN-b
• Bin-packing based algorithm uses the
  first-fit-decreasing-utilization (FFDU)
  heuristic.
• IPC average across all co-schedules
• PART-NOSYM-DYN-e
• Corrects under estimated IPC by adjusting by
  increasing utilization threshold by smallest
  amount
• No task-set is ever rejected
• Simulates the schedule for a hyper-period, to
  determine if it would meet the deadlines.
• Complexity is high, but it gives partitioning the
  fairest showing against global scheduling.

15
Partitioning Algorithms (3)

• PART-NOSYM-STAT
• Independent of co-schedule because of static
  resource allocation
• IPC of each configuration
• FFDU heuristic with an EDF admission test.
• C1, C2, , Cn denote the N hardware contexts.
• Initial all resources are allocated to C1
• Re-allocation resources from C1 to Ck such that
  Ck can accommodate new task
• If C1 dont have enough resource ? Fail
• Remove smallest utilization task form C1 to
  another context that can accommodate it.
• If no such context is found ? Fail

16
Partitioning Algorithms (4)

• PART-SYM-DYN-b
• Utilization average across all co-schedules
• Maximizes average symbiosis among tasks in
  different partitions, while keeping the total
  utilization of tasks in each partition balanced.
• Weighted hypergraph with nodes representing the
  tasks
• A hypergraph is a graph in which generalized
  edges (called hyperedges) may connect more than
  two nodes.
• The weight on a hyperedge (u1, u2, , uN) is the
  inverse of the symbiosis factor of the
  co-schedule formed by tasks u1, u2, , uN.
• Each node is weighted with its tasks
  utilization.
• A hypergraph-partitioning algorithm is used.
• The sum of node-weights (utilization) is
  balanced.
• The weight of the hyperedges is minimized
  (maximizing symbiosis)
• PART-SYM-DYN-e

Reference B. L. Chamberlain. Graph Partitioning
Algorithms for Distributing Workloads of
Parallel Computations
17
Global scheduling algorithms (1)
18
Global scheduling algorithms (2)

• Symbiosis-Oblivious Global Scheduling
• GLOB-NOSYM-PLAIN
• IPC average across all co-schedules
• EDF N tasks with earliest deadlines are chosen
• Tasks with arbitrarily low utilization miss
  deadline (Dhall effect)
• GLOB-NOSYM-US
• EDF-USm/2m-1 algorithm
• If Ti has utilization gt N/(2N-1), give it high
  priority
• Giving the highest priority to high utilization
  tasks in the task set.

Reference A. Srinivasan and S. Baruah.
Deadline-based Scheduling of Periodic Task
Systems on Multiprcessor
19
Global Scheduling Algorithms (3)

• Symbiosis-Aware Global Scheduling
• GLOB-SYM-PLAIN
• Extends EDF to exploit symbiosis in a
  straightforward way.
• It first selects the task with the earliest
  deadline.
• For the other (N-1) tasks, it chooses the set
  that maximizes symbiosis when running with the
  first task.
• Positive ? Improve schedulability (improve
  overall throughput)
• Negative ? Potentially reduce schedulability (no
  real-time characteristic)
• GLOB-SYM-US
• In the presence of high utilization tasks,
  GLOB-SYM-PLAIN impairs schedulability
• Improve the negative of GLOB-SYM-PLAIN
• Defaults to GLOB-NOSYM-US if a task Ti has
  utilization Ui gt N/2N-1
• Otherwise, it defaults to GLOB-SYM-PLAIN

20
Comparison of properties
21
Experiment Setup (1)

• Randomly generated task sets
• Two workloads utilizations follow either normal
  or bi-modal dist.
• of tasks follows an uniform dist. (mean 8)
• Periods from a set 100,200,,1600 with uniform
  probability
• Randomly generated IPCi (mean 3) and co-schedule
  effects on IPCij
• Metric
• Success ratio percentage of tasks successful
  scheduled by an algorithm (at most 5 deadl.
  misses)

22
Experiment Setup (2)
RSIM simulator
Real Workload
23
Results

• Best algorithm ? GLOB-SYM-US
• Partitioning vs. Global algorithm
• Global is generally better
• Enhanced-versions and symbiosis awareness makes
  PART more competitive to GLOB-SYM-US.
• For bimodal, enhanced partition are the best
  (distributes high utilization tasks)
• Symbiosis-awareness
• Partitioning ? often helps
• Global scheduling ? helpful for high
  utilization, not helpful for medium utilization
• PLAIN vs US US does better for bi-modal
  distribution as expected

24
Experimental Methodology

• Metrics
• critical serial utilization (CSU)
• The total utilization obtained by uniformly
  increasing the utilization of all tasks until a
  further increase causes the task-set to become
  unschedulable. (5 deadline ? soft real-time)

25
Results

• Best algorithm ? GLOB-SYM-US
• Static vs. Dynamic resource sharing
• Static resource sharing generally implies lower
  throughput than dynamic resource sharing.
• Partitioning vs. Global algorithm
• Enhanced-version is more competitive to
  GLOB-SYM-US.
• Symbiosis-awareness
• Partitioning ? often helps
• Global scheduling ? it depends

26
Conclusions

• Best algorithm
• Global scheduling, exploits symbiosis,
  prioritizes high utilization tasks, uses dynamic
  resource sharing.
• Require a lot of profiling
• Two alternatives
• Partitioning algorithm that utilizes static
  resource sharing
• (PART-NONSYM-STAT)
• Worse schedulability and somewhat more complex.
• Provide a strict admission control and requires
  less profiling.
• Earliest deadline first global algorithms
• (GLOB-NONSYM-PLAIN)
• Not providing strict admission control, but
  requires no profiling.

27
Conclusions

• Dynamic resource sharing is better than static
  for schedulability
• Partitioning algorithm can be made competitive
  with global scheduling algorithm, but with more
  complexity.
• Symbiosis-awareness
• Beneficial for partitioning algorithms because
  they do not entirely ignore real-time constraint
• Can hurt or help global scheduling algorithms,
  depending on the relative magnitude of the
  symbiosis factors and total utilization of the
  applications.

28

• Thank you