Sunteți pe pagina 1din 55

Intro to HPC and

Supercomputing
Daniel Lucio

Tuesday, October 22, 13


Overview
• What is HPC?
• What is a Supercomputer?
• What is an HPC Center?
• How is it being used?
• The Top500
• Capability vs Capacity
• More information
Tuesday, October 22, 13
What is HPC?
"High-Performance Computing," or HPC, is the
application of "supercomputers" to computational
problems that are either too large for standard
computers or would take too long.

A desktop computer generally has a single processing


chip, limited memory and small storage space, whereas
an HPC system contains thousands of processors, tons
of memory and huge amounts of storage.

Tuesday, October 22, 13


Where did all get this started?
• The CDC 6600 was a
mainframe computer from
Control Data Corporation.

• Delivered in 1964 to the


Lawrence Radiation Laboratory.

• It was used primarily for high-


energy nuclear physics
research, particularly for the
analysis of nuclear events
photographed inside the
Alvarez bubble chamber.

• Had a performance of about


1 megaFLOPS, and remained
the world's fastest computer
from 1964 to 1969.

Tuesday, October 22, 13


What is a
Supercomputer?
• A supercomputer is a computer at the frontline of current
processing capacity.

• Supercomputers are the tool in the field of Computational


Science, and are used for a wide range of computationally
intensive tasks in various fields, including quantum mechanics,
weather forecasting, climate research, oil and gas exploration,
molecular modeling (computing the structures and properties
of chemical compounds, biological macromolecules, polymers,
and crystals), and physical simulations (such as simulation of
airplanes in wind tunnels, simulation of the detonation of
nuclear weapons, and research into nuclear fusion).

Tuesday, October 22, 13


What is a
Supercomputer?
Simulation
“The first great scientific breakthrough of the new
century – the decoding of the human genome
announced in February 2001 – was a triumph of
large- scale computational science.”
(Computational Science: Ensuring America’s Competitiveness, 2005)
“Computational science has become the third
pillar of the scientific enterprise, a peer alongside
theory and physical experiment.”
(Computational Science: Ensuring America’s Competitiveness, 2005)

Nature

Theory Experiment

“A Guide to Monte Carlo Simulations in Statistical Physics”, David Landau, Kurt Binder
Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
Single Multiple
Instruction Instruction
Stream Stream

Single
Data
Stream
SISD MISD
Multiple
Data SIMD MIMD
Stream

Flynn’s Taxonomy is a classification of computer architectures, 1966


Tuesday, October 22, 13
Taxonomy of Computers
SPMD Multiple autonomous processors
simultaneously executing the same
program on different data.

MPMD Multiple autonomous processors


simultaneously operating at least 2
independent programs.

Tuesday, October 22, 13


Taxonomy of Computers
SPMD Multiple autonomous processors
simultaneously executing the same
MPI
program on different data.

MPMD Multiple autonomous processors


simultaneously operating at least 2
independent programs.

Tuesday, October 22, 13


Taxonomy of Computers
SPMD Multiple autonomous processors
simultaneously executing the same
MPI
program on different data.

MPMD Multiple autonomous processors


simultaneously operating at least 2
independent programs.

Tuesday, October 22, 13


Taxonomy of Computers
SPMD Multiple autonomous processors
simultaneously executing the same
MPI
program on different data.

MPMD Multiple autonomous processors


simultaneously operating at least 2
independent programs.

Tuesday, October 22, 13


Taxonomy of Computers
SPMD Multiple autonomous processors
simultaneously executing the same
MPI
program on different data.

MPMD Multiple autonomous processors


simultaneously operating at least 2
independent programs.

Tuesday, October 22, 13


What makes a
Supercomputer?
Processors
& Fast
Memory Interconnect

Software
+
Parallel
File System
Queuing System

Tuesday, October 22, 13


Processors and Memory

Tuesday, October 22, 13


Parallel File System

Tuesday, October 22, 13


Fast Interconnect

Tuesday, October 22, 13


Software
• Operating system

• Scientific applications

• Numerical libraries

• Compilers

• Optimization and Profiling tools

• Visualization tools

• Post-processing tools

• Data transfer software

Tuesday, October 22, 13


Queuing System
There is usually a Load Manager and a Resource Allocator.

Fo r e x a m p l e : T h e re s o u rc e
manager is TORQUE, which
communicates with users
submitting jobs and all of the
compute nodes on the system.
TORQUE monitors memory usage
and processor utilization for all
jobs. TORQUE communicates this
information to Moab. Moab is the
job scheduling program and
determines when and where jobs
can get scheduled.
Tuesday, October 22, 13
Tianhe-2 Supercomputer

#1
Tuesday, October 22, 13
Titan Supercomputer

#2

Tuesday, October 22, 13


Sequoia Supercomputer

#3
Tuesday, October 22, 13
K Supercomputer

Tuesday, October 22, 13


#4
MIRA Supercomputer

#5
Tuesday, October 22, 13
Kraken Supercomputer

#30
Tuesday, October 22, 13
Kraken Specs

• Cray Linux Environment (CLE) 3.1

• A peak performance of 1.17 PetaFLOP Each node has:

• 112,896 compute cores • Two 2.6 GHz six-core AMD


Opteron processors (Istanbul)
• 147 TB of compute memory
• 12 cores
• A 3.3 PB raw parallel file system of disk
storage for scratch space (2.4 PB • 16 GB of memory
available), with capacity of 30 GB/s.
• Connection via Cray SeaStar2+
• 9,408 compute nodes router

• 3D torus interconnect.

Tuesday, October 22, 13


How to use it?

Unix Terminal Science Gateways


A Science Gateway is a
community-developed set of tools,
applications, and data that are
integrated via a portal or a suite of
applications, usually in a graphical
user interface, that is further
customized to meet the needs of a
specific community.

Tuesday, October 22, 13


https://www.xsede.org/web/guest/gateways-listing
Tuesday, October 22, 13
Tuesday, October 22, 13
What is an HPC
Center?
HPC resources People

+
Physical Space Power

Archival Fast networks

Allocations Cooling

Tuesday, October 22, 13


Physical space

Tuesday, October 22, 13


Fast networks

Tuesday, October 22, 13


Fast networks

Tuesday, October 22, 13


Electric power

Tuesday, October 22, 13


Electric power

Tuesday, October 22, 13


Cooling

Tuesday, October 22, 13


Mass storage (archival)
Mass storage facilities often consists of tape and disk
storage components, Linux servers, and High
Performance Storage System (HPSS) software.
Incoming data is written to disk, then later migrated to
tape for long term archival. Tape storage is provided by
robotic tape libraries.

Tuesday, October 22, 13


Allocations

Tuesday, October 22, 13


D.O.E. “HPC” Centers

Tuesday, October 22, 13


N.S.F. “HPC” Centers

Tuesday, October 22, 13


Allocations by Disciplines

INCITE
Tuesday, October 22, 13
What can I do with it?

Tuesday, October 22, 13


What can I do with it?

Tuesday, October 22, 13


Industry needs it too!!!

Tuesday, October 22, 13


• DreamWorks has a "render farm" of servers
made up of about 20,000 processors (HP
BladeSystem c-Class server blades).

• The image rendering jobs are broken up


into small pieces, distributed out to the
server farm, and are later recompiled to
create the final images for a film.

• Required a whopping 80 million compute


hours to render, 15 million more hours than
DreamWorks' last record holder, "The Rise
of the Guardians."

• Between 300 and 400 animators worked on


"The Croods" over the past three years.

• After completing a film, about 70TB worth


of data (things like background art or plants)
is stored for future usage in future
productions.

Tuesday, October 22, 13


The TOP500 list

Tuesday, October 22, 13


How fast is your iPhone5?

1.6GFLOPS
#400 (Nov 1994)

Tuesday, October 22, 13


Capability vs Capacity
Supercomputers generally aim for the maximum in capability computing rather
than capacity computing. Capability computing is typically thought of as using
the maximum computing power to solve a single large problem in the shortest
amount of time.

vs

Google’s Datacenter Kraken Supercomputer

Capacity computing in contrast is typically thought of as using efficient cost-


effective computing power to solve a small number of somewhat large
problems or a large number of small problems, e.g. many user access requests
to a database or a web site.
Tuesday, October 22, 13
What to do with the
results?

Tuesday, October 22, 13


What to do with the
results?

Tuesday, October 22, 13


More information

• http://www.nics.utk.edu
• http://www.xsede.org
• http://www.olcf.ornl.gov/
• http://www.doeleadershipcomputing.org/

Tuesday, October 22, 13

S-ar putea să vă placă și