QoS in Operating Systems Part 3 of Distributed QoS Tutorial

1
QoS in Operating Systems(Part 3 of Distributed
QoS Tutorial)

• Klara Nahrstedt
• Klara_at_cs.uiuc.edu
• http//www.cs.uiuc.edu/klara
• http//cairo.cs.uiuc.edu
• Department of Computer Science
• University of Illinois at Urbana-Champaign

2
Ubiquitous Computing Environment

• Shared Powerful Servers and Thin Clients.
• Intelligence Devices.
• Real Time Response.

Shared PC Clusters SMP machines
Smart Thin Devices Attached to Network
3
Outline

• Motivation and Problem Description
• Reservation-based Systems
• Adaptation-based Systems
• Comparison and Trade-offs

4
Motivation and Problem Description

• Multimedia, ubiquitous and other future computing
  environments consist of variety of hard real-time
  (RT), soft RT and best effort applications
• Need new OS which support coexistence
• Past solutions
• General Purpose OS (UNIX, Windows NT, Linux)
• Serves best effort applications and optimizes
  throughout of schedulable units
• Introduces undesirable gaps and jitters for RT
  applications
• Real-time OS (RT Mach)
• Serves hard RT applications and optimizes timing
  constraints of schedulable units
• Does not provide good throughput and fairness,
  hence best-effort applications may starve

5
Requirements on new OS

• Allow dynamic co-existence and cooperation of
  multiple independently authored RT applications
  with varying timing and resource requirements
• Allow for co-existence of RT and non RT
  applications
• Allow for scheduling of these different
  applications
• Provide protections between various applications

6
Directions

• Two major directions in OS development to
  incorporate QoS
• Reservation-based Systems
• Provide timing guarantees even in overload
  situations
• Adaptation-based Systems
• Provide best possible timing guarantees
• Achieve dynamic reallocation of resources in case
  of scarce resources
• Deliver graceful degradation for QoS-aware
  applications

7
Reservation-based Systems

• Assumptions
• Specification of timing requirements
• Period, Processing Time, Deadline, Burst Duration
• Admission control
• Schedulability tests, bandwidth allocation test
• Enforcement mechanisms to provide requested and
  contracted QoS and allocated corresponding
  resources
• Scheduling algorithms, rate control, buffer
  management

8
Reservation-based Systems

• We will discuss
• Dynamic Soft-Real-Time Scheduling Framework
  (DSRT)
• from University of Illinois at Urbana-Champaign
  by Hao-hua Chu and Klara Nahrstedt (in detail)
• Rialto System
• from Microsoft by Michael Jones and colleagues
• Resource Kernel (RK)
• from CMU by Raj Rajkumar and colleagues

9
Adaptation-based Systems

• Assumptions
• Importance information weight upon which
  metric the resource is allocated
• We will discuss
• Hierarchical CPU Scheduling Framework
• from University of Texas, Austin by Harrick Vin
  et al.
• SMART Framework
• from Stanford University by Jason Nieh and Monica
  Lam