CLUSTERED COMPUTING USING SINGLE THREAD SPACE

[seminar class]

[attachment=11584]
INTRODUCTION
 High performance computing have evolved in past few years.
 Their taxonomy is based upon processors, memory and interconnect.
TYPES OF COMMON SYSTEMS
 Massively Parallel Processors (MPP).
 Symmetric Multiprocessors (SMP).
 Cache-Coherent Non Uniform Memory Access (CC-NUMA).
 Distributed Systems.
 Clusters
CLUSTERS
 Based on interconnection network.
 Computers run integration software to support collaborative computing.
 Hides physical boundaries and appears as a Single System Image (SSI).
SINGLE THREAD SPACE
 Threads freely move between machines
 Cluster as a new environment for multithreaded applications.
 Threads running within this space would share all the resources that the thread has created.
 Provides single system illusion among all the threads created by a process.
DESIGN
 The Delta Execution subsystem
 Master Slave Redirection subsystem
A Distributed shared memory (DSM)
DELTA EXECUTION
 A preemptive thread mechanism.
 Supports transparent thread to processor mapping within the single thread space.
 Provides high level and portable implementation of java thread.
 Completely hides all the system dependent details.
GLOBAL OBJECT SPACE
 Provides location independent object access.
 Creates a globally accessible object space.
 Employs a decentralized approach for the memory management.