High Performance Computing and TupleSpaces

1
High Performance Computing and TupleSpaces

• Student Jeremy Kemp
• Advisor Hong Lin

2
High Performance Computing

• Our cluster Grid
• Rocks 5.1 64-bit
• What TSpaces uses

3
What is Grid?
4
What is Grid?

• Cabinet 0

• Cabinet 1

• 17 Nodes
• 3.06 GHz
• 1GB RAM

• 17 Nodes
• 1.8 GHz
• 2GB RAM

• Images courtesy of Intel.com

5
What is Grid?

• Head Node Grid itself

• Dual Socket Quad Core Xeon Workstation
• 8 cores total
• Jobs are Started Here
• Hosts all the data (for now)
• 12GB of RAM

• Images courtesy of Intel.com

6
What is Grid?
7
What is Rocks?

• Developed at UCSD using an NSF grant
• It allows a cluster to be installed and running
  in under a day
• Comes with Cluster infrastructure tools
• Contains all of the things a cluster needs

8
What does TSpaces use?

• Network File System
• Pre-configured public key authentication
• Automatic user account creation
• DNS allows easy access to the same TSpaces Server

9
Tspaces

• TSpaces is a Java package that contains
• Tspaces server
• Java module containing methods and data
  structures used in programming a tspaces program
• Used primarily as a cross-platform communications
  platform

10
TupleSpaces

• This is what TSpaces is named after
• Runs on one host
• All data is stored here
• Programmer uses TSpace methods to access
• Full of Tuples

11
The Tuple

• All data stored in a TSpace must be inside a
  Tuple
• Can hold primitives and classes, including user
  defined classes
• Tuples can be nested
• Tuples are a subclass of superTuple

12
Tspaces Methods

• read()
• take()
• waitToread()
• waitToTake()
• write()
• countN()
• delete()
• deleteAll()

13
Configuring TSpaces

• Unpack the tarball to the NFS
• Using GNU Screen, open a persistent instance of
  the TSpaces server
• Add the location of the TSpaces files to the
  user's classpath
• Run the program on the number of nodes desired

14
GammaSort

• An Example program modeled after the Chemical
  Reaction Model
• Is parallelizable to n/2 nodes
• 4 Other programs written to debug/run gammaSort
• dataGenerator.java
• dataViewer.java
• dataCleaner.java
• dataReset.java

15
GammaSort

• Data elements consist of a tuple
• tuple dataTemplate new Tuple(new
  Field(Integer.class), new Field(Double.class))
• The integer is the index of the search
• The above is the template, used to 'search' for
  the desired tuple using read() and take()

16
GammaSort

• Data elements are pulled from the TupleSpace by
  the compute node and compared
• If they are out of order they are switched
• The Data elements are then written back to a
  tuplespace

17
GamaSort and Synchronization

• In order to synchronize the data, 4 counters are
  used
• 4 tupleSpaces are used
• Two TupleSpaces for the data elements being read
  (one even and one odd)
• Two TupleSpaces for the data elements being
  written (one even and one odd)
• All data must be handled in pairs to determine
  when to Terminate the program.

18
Future Improvements

• Consolidate all of the support applications in to
  one, customizable GUI application to help build
  future programs in TSpaces
• Use more of the Features and Customizations to
  provided by TSpaces
• Automated Startup tool/script