What is a Computer Cluster?
• A computer cluster is a group of computers
working together as a single system. Each
node performs the same task, managed by
specialized software. Unlike grid computing,
all nodes in a cluster typically run the same
application.
Example: Sun Microsystems Solaris
Cluster
• Sun Microsystems developed clusters using its
Solaris operating system for enterprise-level
parallel processing.
Cluster Components and
Connections
• Nodes are connected via high-speed LANs and
run their own OS instances. Each computer
(node) acts as a server in the cluster.
Hardware and Software Uniformity
• Clusters typically use identical hardware and
OS. However, systems like OSCAR allow
heterogeneous setups. OSCAR is Linux-based
for managing high-performance clusters.
Benefits of Clusters
• • Improved performance
• • Increased availability
• • Cost-effective compared to supercomputers
Why Clusters Became Popular
• Driven by:
• • Cheap microprocessors
• • Fast networking
• • Readily available software for distributed
computing
Clusters vs. Mainframes
• Clusters are cheaper and easier to scale than
fault-tolerant mainframes, but managing
failures is more complex.
Challenges: Administration
• Managing N-node clusters can be as expensive
as managing N independent machines.
Challenges: Task Scheduling
• In heterogeneous environments, scheduling
tasks efficiently is difficult. Performance
depends on job and hardware characteristics.
Challenges: Task Mapping
• Mapping tasks to CPU cores and GPU devices
is complex and critical for performance.
Challenges: Node Failure
• Use fencing to isolate or restrict
malfunctioning nodes. Two types:
• 1. Disable node
• 2. Block access to shared resources.
Linux Cluster Software
• • distcc – distributes compilation
• • MPICH – for message passing
• • Linux Virtual Server, Linux-HA – manage
service distribution
Kernel-Level Clusters in Linux
• • MOSIX, Kerrighed, OpenSSI – full integration
• • Enable automatic process migration
• • Single-system image for users
Windows Cluster Server 2003
• Microsoft’s clustering includes:
• • Job Scheduler
• • MSMPI library
• • Management tools for high performance
computing

Computer_Cluster in Parallel and Distributed computer.pptx

  • 1.
    What is aComputer Cluster? • A computer cluster is a group of computers working together as a single system. Each node performs the same task, managed by specialized software. Unlike grid computing, all nodes in a cluster typically run the same application.
  • 2.
    Example: Sun MicrosystemsSolaris Cluster • Sun Microsystems developed clusters using its Solaris operating system for enterprise-level parallel processing.
  • 3.
    Cluster Components and Connections •Nodes are connected via high-speed LANs and run their own OS instances. Each computer (node) acts as a server in the cluster.
  • 4.
    Hardware and SoftwareUniformity • Clusters typically use identical hardware and OS. However, systems like OSCAR allow heterogeneous setups. OSCAR is Linux-based for managing high-performance clusters.
  • 5.
    Benefits of Clusters •• Improved performance • • Increased availability • • Cost-effective compared to supercomputers
  • 6.
    Why Clusters BecamePopular • Driven by: • • Cheap microprocessors • • Fast networking • • Readily available software for distributed computing
  • 7.
    Clusters vs. Mainframes •Clusters are cheaper and easier to scale than fault-tolerant mainframes, but managing failures is more complex.
  • 8.
    Challenges: Administration • ManagingN-node clusters can be as expensive as managing N independent machines.
  • 9.
    Challenges: Task Scheduling •In heterogeneous environments, scheduling tasks efficiently is difficult. Performance depends on job and hardware characteristics.
  • 10.
    Challenges: Task Mapping •Mapping tasks to CPU cores and GPU devices is complex and critical for performance.
  • 11.
    Challenges: Node Failure •Use fencing to isolate or restrict malfunctioning nodes. Two types: • 1. Disable node • 2. Block access to shared resources.
  • 12.
    Linux Cluster Software •• distcc – distributes compilation • • MPICH – for message passing • • Linux Virtual Server, Linux-HA – manage service distribution
  • 13.
    Kernel-Level Clusters inLinux • • MOSIX, Kerrighed, OpenSSI – full integration • • Enable automatic process migration • • Single-system image for users
  • 14.
    Windows Cluster Server2003 • Microsoft’s clustering includes: • • Job Scheduler • • MSMPI library • • Management tools for high performance computing