Scalability In Distributed Systems

1
Scalability In Distributed Systems

• DISTRIBUTED SYSTEMS CS 775
• PHANI DEEP NARRA
• pnarra_at_cs.odu.edu

2
CONTENTS

• INTRODUCTION
• GOALS OF A DISTRIBUTED SYSTEM
• SCALABILITY INTRODUCTION
• SCALABILITY PROBLEMS
• SCALABILITY SOLUTIONS
• CONCLUSION
• REFERENCES
• QUESTIONARIES

3
INTRODUCTION

• What is a Distributed System?
• - A collection of interconnected computers
• Characteristics of a Distributed System
• - Limited Bandwidth
• - Latency
• - Failure Threats

4
GOALS OF A DISTRIBUTED SYSTEM

• Heterogeneity
• Concurrency
• Transparency
• Openness
• Security
• Reliability
• Scalability

5
SCALABILITY

• What is scalability?
• - A change in the quantity of resources
  available
• The three Dimensions of Scalability
• - Size
• - Geographical Coverage
• - Administration

6
SCALABILITY PROBLEMS

• Increase in the size of a distributed system in
  terms of the number of resources and the number
  of users in the system.
• Limitations on scalability in size
• - Centralized Servers
• - Centralized Data
• - Centralized algorithms

7
SCALABILITY PROBLEMS

• Centralized Servers
• A distributed system with a single server
• The server becomes a bottleneck as the number of
  users increase.
• Increase the number of servers in the system
• It is sometimes unavoidable
• Ex Bank accounts

8
SCALABILITY PROBLEMS

• Centralized Data
• Centralized Data are as bad as Centralized
  Servers.
• Communication links get saturated with
  centralized Data(Ex DNS)
• Have data distributed to many servers.

9
SCALABILITY PROBLEMS

• Centralized algorithms
• Centralized algorithms is also a bad idea.
• Every node(computer) has the information about
  the state of the system.
• In large DSs number of messages are to be routed
  between nodes(computers).
• Information about the load on all the nodes is
  collected.

10
SCALABILITY PROBLEMS

• Information is sent as messages to all nodes.
• These messages overload the system.
• Solution
• Use of Decentralized Algorithms with the
  following characteristics
• No machine has complete information about the
  system state.
• Decisions taken based on local information.
• Failure of a node doesnt ruin the algorithm.
  

11
SCALABILITY PROBLEMS

• Ability to perform when the distance between the
  users increases.
• It was hard to geographically scale a LAN system
  into a WAN system.
• Reason being Synchronous communication.
• Client blocks itself until reply received.
• Moreover the communication in wide area networks
  is highly unreliable.
• Geographical Scalability is strongly related to
  Size reliability.

12
SCALABILITY PROBLEMS

• The ability for an increasing number of
  independent administrative domains to easily
  share a single distributed system.
• Many components of a distributed system that
  generally reside within an administrative domain
  operate in their domain.
• In that case, system administrators might have
  tested and certified the applications and the
  users of that component trust it.
• However, this trust does not go beyond the
  administration boundaries.

13
SCALABILITY PROBLEMS

• If it expands to another domain, two security
  issues need to be considered.
• First, the DS has to protect itself from
  malicious attacks from the new domain.
• Second, the new domain has to protect itself from
  the malicious attacks from the Distributed System.

14
SCALABILITY SOLUTIONS

• There are three basic techniques for scaling
• Hiding communication latencies
• Distributing the Services
• Replicating the data
• Hiding communication latencies is applicable to
  Geographical scaling.
• Try to avoid waiting for responses and do some
  other useful work until response received.

15
SCALABILITY SOLUTIONS

• There are many systems that cannot use
  asynchronous communication.
• Distribution technique can be used as a solution
  to size scalability.
• Distribution means taking a component, splitting
  it into parts and spreading those parts among the
  nodes of the system(Ex DNS)
• Replication of data can be used for size
  scalability.

16
CONCLUSION

• A distributed system must be designed such that
  it is scalable.
• We need to face all the problems discussed thus
  far and come up with solutions for them in order
  to build a successful long lasting distributed
  system.
• Once again, the problems discussed here are a
  single drop in an ocean full of problems.

17
REFERENCES

• http//en.wikipedia.org/wiki/Scalability
• http//ieeexplore.ieee.org
• Reliable Distributed Systems, Technologies, Web
  Services and Applications by Kenneth P Birman.
• Design, performance and scalability of the
  distributed enterprise systems, by Janusz
  Kowalik.
• Distributed systems principles and paradigms
  by Andrew S. Tenanbaum.

18
Questions?