Cluster Computing in the Classroom: Topics, Guidelines, and Experiences

1
Cluster Computing in the Classroom Topics,
Guidelines, and Experiences

• Amy Apon
• Department of Computer Science Computer
  Engineering
• University of Arkansas

2
Clusters and Data Engineering

• A cluster is a set of whole computers connected
  via a network, and used as an integrated resource
  to solve a single application
• Increase throughput for massive data processing
• Inexpensive - uses commodity computers with lots
  of disks and disk space

3
Teaching Challenges

• Prerequisites are difficult to establish
• One course does not fit all!

We propose

• Cluster teaching material organized as modules
• Accessible to a variety of situations

4
Outline

• Overview of target audience for the proposed
  teaching materials
• Description of course modules
• Problem areas
• Conclusions
• Acknowledgements and references

5
Courseware developed with
Dr. Amy Apon
Dr. Jens Mache
Dr. Hai Jin
Dr. Rajkumar Buyya
6
Who our students are

• Juniors, seniors, graduate students
• With a variety of preparation
• Operating Systems?
• Maybe havent seen threads
• Computer Networks?
• Maybe havent seen sockets
• Computer Architecture?
• Maybe dont understand how cache works

7
Course Units

• Needed because of the diversity of institutions
  and student preparation
• Matched to the Computing Curricula 2001 to avoid
  overlap with existing courses
• Basic Units (have overlap with ACM Core)
• Core Units (essential to cluster computing)
• Extended Units (more advanced, optional)

8
Course Units Can Be Combined

• We propose sample courses with an emphasis in one
  of
• Architecture
• Programming
• Algorithms and Applications

9
Five Basic Units

• Programming Fundamentals (PF2, PF5)
• Algorithms and problem-solving
• Event-driven programming (3 hours total)
• Architecture and Organization 4 (AR4)
• Memory system organization (1 hour)
• Architecture and Organization 7 (AR7)
• Multiprocessing architectures (1 hour)
• Operating Systems 3 (OS3)
• Concurrency (1 hour)
• Net-Centric Computing 2 (NC2)
• Communication and networking (2 hours)

10
Ten Core Units

• Algorithms and Complexity 4 (AL4)
• Distributed algorithms (1 hour)
• Algorithms and Complexity 11 (AL11)
• Parallel algorithms (3 to 7 hours)
• Architecture and Organization 7 (AR7)
• Multiprocessing and alternative architectures (2
  hours)
• Architecture and Organization 9 (AR9)
• Architectures for networks distributed systems
  (1-4 hours)
• Operating Systems 11 (OS11)
• System performance evaluation (1-2 hours)

11
Ten Core Units, continued

• Net-Centric Computing 2 (NC2)
• Communication and networking (1 hour)
• Net-Centric Computing 6 (NC6)
• Network management (1-2 hours)
• Social and Professional Issues 9 (SP9)
• Economic issues in computing (2 hours)
• Software Engineering 2 (SE2)
• Using APIs Basic MPI or PVM, basic PVFS (2
  hours)
• Computational Science 4 (CN4)
• High-performance computing (6 or more hours)

12
Many Choices for Extended Units!

• Software Engineering (SE3), Software tools and
  environments
• Debugging tools
• Operating Systems (OS8)
• Parallel file systems
• Algorithms (AL11)
• Advanced parallel algorithms.
• Architecture and Organization (AR9)
• Architecture for networks and distributed systems
• Graphics and Visualization (GV9)
• Intelligent Systems 4 (IS4), Advanced search
• Information Management (IM8, IM9, IM10, IM11)
• Distributed databases, physical database design,
  data mining, and information storage and
  retrieval on clusters
• Computational Science (CN1, CN3)

13
Cluster Architecture Emphasis

• Similar requirements as for a course in advanced
  computer architecture
• Suited for advanced undergraduates and graduate
  students who have completed
• Computer organization
• Computer networks
• Operating systems
• Programming

14
Cluster Architecture Topics
15
Programming Emphasis

• Suited for undergraduates with exposure to
• Data structures and algorithms
• Computer organization
• Can use general access computer lab/LAN (if
  performance is not an issue)
• Can use generally available programming
  environments

16
Cluster Programming Topics

• Shared memory programming
• Leading to a discussion of NUMA
• Sockets
• Leading to discussion about network overhead,
  low-latency protocols
• Parallel programming using MPI
• Middleware Java RMI, CORBA

17
Algorithms and Applications

• Suited for
• Advanced undergraduate with a strong algorithms
  and programming background
• Graduate students
• Can be
• Parallel algorithms
• With a focus on topics from a particular domain

18
Algorithms and Applications Topics

• Application Overview
• Compression, data mining, image rendering,
  genetic algorithms,
• Techniques of Algorithm design
• Partitioning, divide and conquer, communication
  and synchronization,
• Modeling and visualization
• Performance tuning

19
Classroom Favorites

• Build your own cluster
• Using old lab machines, install PVM or MPI
• Parallel matrix multiply, sort
• Implement these using MPI, evaluate the
  performance using data of varying size, present
  results graphically
• Term programming project
• Can have students select their own!

20
Problem Areas

• Cluster setup and administration
• Cluster usage (especially for performance
  experiments)
• Security

21
Conclusions

• Cluster computing is a low cost approach to
  massive data processing
• Cluster computing can be taught at the
  undergraduate level
• Modules help to organize the material so that it
  is appropriate for your institution
• Modules can be mixed and matched

22
References and Acknowledgements

• Cluster Computing in the Classroom Topics,
  Guidelines, and Experiences 
• by Amy Apon, Rajkumar Buyya, Hai Jin, Jens Mache,
  First International Workshop on Cluster Computing
  Education, Cluster.Edu 2001
• See http//citeseer.nj.nec.com/395286.html