HPC and HTC Concepts

Key Concepts

High Performance Computing (HPC)

High Throughput Computing (HTC)

AKA "Supercomputing"

The Computer

Actual Rackable Server

Faster computing

• Reduce time to result

• Increase accuracy

• Increase throughput

Faster computing

• Pressure from computing requirements

• Pressure from physics

Pressure from Physics

Physics limits GHz speed

Heat and power-draw increase super-linearly with speed

Pressure from computing requirements

• Bottlenecks everywhere

• Limits of RAM

Parallelism

• Serial Process:

  • A process in which its sub-processes happen sequentially in time.
  • Only one sub-process is active at any given time.

• Parallel Process:

  • Process in which multiple sub-processes can be active simultaneously.

Parallelism

• Serial Process:

  • Speed depends only on the rate at which each sub-process will occur (e.g. processing unit clock speed).

• Parallel Process:

  • Speed depends on execution rate of each sub-process AND how many sub-processes can be made to occur simultaneously.

Parallel for Capacity

Example: Searching database of web-sites for some text (i.e. Google)

• Searching sequentially through large volumes of text too time-consuming.

  • Multiple servers hold different pages.
  • Each server can report the result of each individual search.
  • The more servers you add the quicker the search is.

Parallel for Capability

Example: Large scale weather simulation

• Detailed description for atmosphere too large to run on today's desktop or server PCs.

  • Multiple servers are needed to hold all grid data in memory.
  • Servers need to quickly communicate to synchronise work over entire grid
  • Communication between servers can become a bottleneck.

Parallelism

All computers today are designed for parallel execution

But...

Not all computation can be done in parallel!

Amdahl's Law

1 Processor

Amdahl's Law

2 Processors

Amdahl's Law

10 Processors

Amdahl's Law

• Amdahl's law will always be a limiting factor

$\frac{1}{\left(1-P\right)+\frac{P}{N}}$

• Some applications need to be re-written, algorithms re-thought to harness ever more parallel machines.

Amdahl's Law

• If 95% of the program can be parallelized, the theoretical maximum speedup using parallel computing would be 20x, no matter how many processors are used.

There are many parallel paradigms

Processor farms, pipelining, divide/conquer, geometric decomposition, cellular automata, algorithmic parallelism

• From a systems perspective:
  • Shared memory
  • Message passing

SMP (Symmetric Multi-Processing)

SMP (Symmetric Multi-Processing)

Beowulf Cluster

Infiniband Cluster

NUMA (Non-Uniform Memory Access)

Interconnect comparisons

Interconnect	Typical MPI latency (microseconds)	Typical bandwidth (MB/s)
1Gbs Ethernet	60-90	90
10Gbs Ethernet	12-20	800
Infiniband	2-4	250-1200
NUMALink 4	~1	3000
QPI	~0.5	20000

GPU accelerator

Only a subset of problems can be optimised

• Not all problems can be parallelised

• Not all parallel problems can be ported to accelerators

In summary

• Developments in CPU architecture still continue: Moore’s law still valid

• Improvements to CPU architecture consist of increasing internal parallelism: hyper-threading, wider SIMD units, more cores per chip

• Accelerators are bringing more computing cycles to the table (albeit of a special type).

And that is just the compute side of the story

Data storage and Network infrastructure need to keep up with computation and other data producing technologies...

• upgrading network infrastructure

• re-thinking data storage

  • SSD’s
  • parallel file systems
  • Multi-tiered storage
  • Map-Reduce data platforms

Supercomputing

Used efficiently, Supercomputers let you get more done faster.

They can be useful for a large number of types of work.

Applications

The Cluster

Overview

The Node

• Compute Node

  • Just computes - little else
  • Private IP address - no user access

• Login Node

  • User login
  • Interaction with job scheduler
  • Public IP address - connects to external network

Generic Software stack

Real applications

• FEA - Finite Element Analysis

  • The simulation of hard physical materials, e.g. metal, plastic
  • Crash test, product design, suitability for purpose
  • Examples: MSC Nastran, Ansys, LS-Dyna, Abaqus, ESI PAMCrash, Radioss

• CFD - Computational Fluid Dynamics

  • The simulation of soft physical materials, gases and fluids
  • Engine design, airflow, oil reservoir modelling
  • Examples: Fluent, Star-CD, CFX

Real applications

• Geophysical Sciences

  • Seismic Imaging - taking echo traces and building a picture of the sub-earth geology
  • Reservoir Simulation - CFD specific to oil asset management
  • Examples: Omega, Landmark VIP and Pro/Max, Geoquest Eclipse

• Life Sciences

  • Understanding the living world - genome matching, protein folding, drug design, bioinformatics, organic chemistry
  • Examples: BLAST, Gaussian, LAMMPS, Trinity, Amber, NAMD

Real applications

• High Energy Physics

  • Understanding the atomic and sub-atomic world
  • Software from Fermi-Lab or CERN, or home-grown

• Financial Modelling

  • Meeting internal and external financial targets particularly regarding investment positions
  • VaR - Value at Risk - assessing the impact of economic and political factors on the bank’s investment portfolio
  • Trader Risk Analysis - what is the risk on a trader’s position, a group of traders

Operating system

• The vast majority of Clusters in the world use some flavour of Unix or Linux for their OS.

• The most common form of interaction with these systems is a "shell" or "command line".

(Which we are going to learn about using Legion’s Login nodes.)

Operating system

• Clusters are very frequently used as a shared facility.

• As such, work needs to be scheduled via a batch system.

• Jobs are queued and prioritised based on requested resources.

(These are the focus of tomorrow’s session.)