HPC(High Performance Computing)

Unit 1 – Cluster Computing at Glance

What is Cluster Computing?
 A computer cluster consists of a set of loosely or tightly connected computers that work
together so that, in many respects, they can be viewed as a single system.
 Unlike grid computers, computer clusters have each node set to perform the same task,
controlled and scheduled by software.

1. Scalable Parallel Computer Architectures
 In past decades many high performance computing have emerged.
 Their classification is based on how their processors, memory and interconnect are laid out.
 The most common parallel computing architectures are as following
o Massively parallel processors (MPP)
o Symmetric multiprocessors (SMP)
o Cache-coherent non-uniform memory access (CC-NUMA)
o Distributed Systems
o Clusters




Table 1.1 Key characteristics of Scalable parallel computing architectures


2. A Cluster Computer and its Architecture
 A cluster is a type of parallel or distributed processing system.
 It consists of a collection of interconnected stand-alone computers working together as a single,
integrated computing resource.

Ishan K. Rajani, CE Department | 170704 – Advance Computing Technology 1

, Unit 1 – Cluster Computing at Glance

 It is a collection of computers or workstations working together as a single, integrated
computing resource connected via high speed interconnects.




Figure 1.1 Cluster computer architecture

Followings are the components of a cluster computer:

 High performance computers: These can be PCs, workstations, SMP. These nodes generally
have memory, I/O facility and operating system.
 Modern operating system: OS should be layered or micro kernel based. Typical OS used in
clusters are Linux, Solaris, Windows NT.
 High speed networks/Switches: Interconnects as Gigabit Ethernet, ATM etc are suitable and
high performance switches are used for connections.
 Network Interface hardware is generally various NICs which is an interface to network.
 Communication Software includes various fast communication protocols and services as AM,
FM which is used for speedy communication between nodes.
 Cluster Middleware: its basic aim is to provide SSI and SAI services to the user. It includes
o Hardware like Hardware DSM, Memory channel etc.
o Operating system kernel also called the gluing layer like Solaris MC, GLUnix etc.
o Applications and subsystem like Software DSM, File systems Management tools, RMS
software etc
 Parallel Programming environment and Tools: These provide portable, efficient and easy to
use tools to work in parallel environment and run applications simultaneously. E.g. PVM, MPI
etc.
 Applications: Sequential as well as parallel applications can be run on a cluster.

Ishan K. Rajani, CE Department | 170704 – Advance Computing Technology 2

, Unit 1 – Cluster Computing at Glance

3. Cluster Classification
 Clusters are classified into many categories based on various factors as indicated below.

 Application target
o Computational science or mission-critical applications.
o High performance(HP) clusters
o High Availability(HA) clusters

 Node ownership
o Owned by individual or dedicated as a cluster node.
o Dedicated clusters
o Non-dedicated clusters

 Node hardware
o PC, Workstation, or SMP
o Clusters of PCs (CoPs) or Piles of PCs (PoPs)
o Clusters of Workstations (COWs)
o Clusters of SMPs (CLUMPs)

 Node operating system
o Linux, NT, Solaris, AIX etc.
o Linux Clusters (Ex: Beowulf)
o Solaris Clusters (Ex: Berkley NOW)
o NT Clusters (Ex: HPVM)
o AIX Clusters (Ex: IBM SP2)
o Digital VMS Clusters
o HP-UX Clusters
o Microsoft wolfpack clusters

 Node configuration
o Node architecture and type of OS it is loaded with.
o Homogeneous clusters: Simil
o Heterogeneous clusters

 Levels of clustering
o Based on location of nodes and their count
o Group clusters: Nodes are connected through SAN (System area networks)
o Departmental clusters: 10s to 100s nodes
o Organizational clusters: more than 100s node
o National meta-computers (WAN/Internet-based)
o International meta-computers (Internet-based)




Ishan K. Rajani, CE Department | 170704 – Advance Computing Technology 3

, Unit 1 – Cluster Computing at Glance

4. Commodity Components for Clusters
 Followings are the components which are commonly used to build clusters and nodes.

 Processors
o Computer systems based on these processors have been used to form clusters.
o Fo e a ple, Be kele NOW uses Su s SPA‘C fa il of p o esso s i thei luste
nodes.
o Other popular examples of processors includes x86 variants (AMD x86, Cyrix x86), Digital
Alpha IBM PowerPC, Sun SPARC, SGI MIPS, and HP PA.

 Memory and Cache
o Nodes can use various types of memory and they include extended data out(EDO) and
fast page.
o EDO allows the next access to begin while previous data is still being read.
o Fast page allows multiple adjacent accesses to be made more efficiently.
o Caches are used to keep recently used blocks of memory for very fast access if the CPU
references a word from that block again.

 Disk and I/O
o One way of improving I/O performance is to carry out I/O operations in parallel.
o It can be done by parallel file systems based on hardware or software RAID.
o Hardware RAIDs can be expensive, Software RAID can be constructed by using disks
associated with each workstation in the cluster.

 System Bus
o The initial PC bus known as ISA bus bas worked with 16 bits wide and clocked in excess
of 13 MHz.
o It was not sufficient to meet the demand of the latest CPUs, disk interfaces and other
peripherals.
o Group of PC manufacturers introduced the VESA local bus, 32-bit bus that matched the
s ste s lo k speed.

 Cluster Interconnects
o Ethernet, Fast Ethernet, and Gigabit Ethernet
o Asynchronous transfer mode (ATM)
o Scalable coherent interface (SCI)

 Operating systems
o LINUX
o Solaris
o Microsoft window NT




Ishan K. Rajani, CE Department | 170704 – Advance Computing Technology 4