Distributed System Design Goals

1
Section 2.1

• Distributed System Design Goals
• Alex De Ruiter
• aderuiter1_at_student.gsu.edu

2
Why a Distributed System?

• Resource sharing
• Specialized hardware
• Data/Database
• Computation speedup
• Partitioning computations allowing concurrent
  processing
• Load balancing
• Reliability
• In event of system resource failure, shift its
  processing load to another similar resource.

3
Operating System Design Goals

• Efficiency
• Flexibility
• Consistency
• Robustness

4
Efficiency

• Standard measurements
• Throughput Number of process/tasks completed per
  unit of time.
• Turnaround Time Process/task specific time
  required to execute from start to finish.
• Waiting Time Time process spent waiting to
  execute.
• Response Time If in an interactive environment,
  the time required to provide the first response
  to a processing request.

5
Efficiency

• Distributed system efficiency issues
• Data propagation
• In a centralized system, access to data is for
  the most part immediate. Not so the distributed
  system, data may be found in widely dispersed
  locations.
• Solution,
• Data migration
• Computation migration
• Process migration

6
Efficiency

• Communication protocol overhead
• Once again, in a centralized system a common
  basis for communication is present. Not so the
  distributed system.
• Differing character encoding- ASCII vs EBCDIC
• Differing structuring of binary data.
• Big Endian vs Little Endian
• In a centralized system no need for network
  protocols to facilitate interprocess
  communication.
• Solutions/requirements
• Tightly coded communication constructs at the
  operating system level. Handle the ASCII vs
  EBCDIC or Endian issues with a minimum of fuss.
• Effective communication protocols at the network
  level

7
Efficiency

• Load distribution
• Issues bottlenecks and congestion
• Network
• WAN
• LAN
• Software components / processes
• Scalability also falls in the robustness
  category

8
Flexibility

• Friendliness, what is it?
• ease of use of the system interface
• Ability to relate computation processes to the
  users problem space - accountability
• Freedom, what is it?
• The ability to select when, where and how to use
  the system with restrictions being at a minimum.
• Extensibility of the user environment. Ready
  system features present to provide for the
  creation of new user tools and services.

9
Consistency

• What is it?
• Uniformity in using the system.
• Predictability in system behavior.
• Issues
• Lack of global information
• Timing
• System state
• Component failures
• Data replication and partitioning
• Maintaining data integrity in the face of
  concurrency over widely dispersed data locations

10
Robustness

• A system is said to be robust if it is fault-
  tolerant. Or said another way, it has the
  ability to continue to function, perhaps in a
  reduced capacity, after the occurrence of some
  form of system failure.
• Common distributed system failures
• Communication link failure
• Processing nodes failure / site outage
• Client/server processes failure / message loss

11
Robustness

• Recovery, step one detect the failure.
• Detect that some system resource has ceased to
  function.
• Recovery, step two reconfiguration.
• Isolate the failed system resource from the
  system as a whole to prevent attempted use of
  said failed resource.
• Recovery, step three restoration.
• Once the failed resource returns to operation the
  system must be once again reconfigured to
  include the formerly failed resource.

12
Robustness

• Scalability can also be considered under the
  heading of robustness.
• Scalability is the ability of a system to adapt
  to an increased service load.
• Any system resource who's load is proportional to
  the size of the system has the potential of being
  overwhelmed as the system increases in size.
• So, scalability, and therefore robustness,
  requires that the maximum load on any system
  resource be determined by some value other than
  system size.

13
Robustness

• Security
• Breach of confidentiality - Access of data
• Breach of integrity - Modification of data
• Breach of availability - Destruction of data
• Theft of service - Use of resources
• Denial of Serice - Preventing resource use
• Protection
• Allow multiple users/processes access to shared
  system resources in a manner consistent with
  system protection polices.
• i.e. only allow a user/process to access the
  minimum set of resources needed to accomplish
  their task.

14
Questions?
15
References

• Randy Chow, Theodore Johnson, Distributed
  Operating Systems Algorithms, Addison Wesley,
  pp.28-29.
• Abraham Silberschatz, Peter Baer Galvin, Greg
  Gagne, Operating System Concepts, John Wiley
  Sons Inc., pp. 612 - 617