Seminario Paolo Maggi, 24-05-2012

Tecnologie Cloud per la fornitura di applicazioni
ambientali HPC in modalità SaaS

Paolo Maggi <paolo.maggi@nice-software.com>
R&D Manager

Copyright © NICE - 2012

Tecnologie Cloud per la fornitura di applicazioni
ambientali HPC in modalità SaaS

Software
High
as Performance
a Computing
Service


What is High Performance Computing?

Refers to the use of supercomputers or clusters of
computers to solve difficult computational problems that
typically arise through scientific inquiry.
Computer users turn to HPC when a problem is too
large to solve on a conventional laptop or workstation or
runs too slowly
– it requires too much memory or disk space
– the algorithm is complex
– the dataset is large
– data access is slow


When we need HPC?

To do a time-consuming operation in less time
– I am an aircraft engineer
– I need to run a simulation to test the stability of the wings at high
speed
– I’d rather have the result in 5 minutes than in 5 days so that I
can complete the aircraft final design sooner


When we need HPC?

To do an operation before a tighter deadline
– I am a weather prediction agency
– I am getting input from weather stations/sensors
– I’d like to make the forecast for tomorrow before tomorrow


When we need HPC?

To do a high number of operations per seconds
– I am an engineer of ACME.com
– My Web server gets 1,000 hits per seconds
– I’d like my web server and my databases to handle 1,000
transactions per seconds so that customers do not experience
bad delays


How to get my answer faster?

Work harder
– Get faster hardware

Work smarter
– Use optimized algorithms (libraries!)
– Write faster code (adapt to match hardware)
– Trade convenience for performance (e.g. compiled program vs.
script program)

Delegate parts of the work
– Parallelize code
– Use accelerators (GPGPU/CUDA)


Leveraging parallelism



Functional parallelism
Different people are performing different tasks at the same time



Data parallelism
Different people are performing the same task, but on different (equivalent) objects


What a cluster is…

A cluster needs:
– Several computers, nodes, often in special cases for easy
mounting in a rack
– One or more networks (interconnections) to hook the nodes
together
– Software that allows the nodes to communicate with each other
(e.g. MPI)
– Software, called Distributed Resource Manager (DRM), that
manage resources and assign them to individual users and
computational jobs (e.g. LSF, PBS, SGE, …)

A cluster is: all of those components working together to
form one big computer (to give you answers in a fast way)


How a cluster looks like?


What is Cloud Computing?


A working definition of Cloud Computing (NIST)

Cloud computing is a model for enabling convenient,
on-demand network access to a shared pool of
configurable computing resources (e.g., networks,
servers, storage, applications, and services) that can
be rapidly provisioned and released with minimal
management effort or service provider interaction.
This cloud model promotes availability
Composed of
– five essential characteristics
– three service models
– four deployment models


The NIST Cloud definition framework

On Demand Self-Service
Essential
Broad Network Access Rapid Elasticity
Characteristics
Resource Pooling Measured Service

Service Software as a Platform as a Infrastructure as a
Models Service (SaaS) Service (PaaS) Service (IaaS)

Hybrid Clouds
Deployment
Models Private Community
Public Cloud
Cloud Cloud


Cloud Computing Offerings


About NICE
Pioneer in Technical and
Engineering Cloud solutions
14 years experience with enterprise
Grid/Cloud solutions throughout all
industries
Worked closely with ISV and HW vendors
Headquarters: Italy since the beginning
Offices: USA, Germany Strong relationship with Research and
Academia

Core business: Access to Grid / HPC
solutions and Remote Visualization
EnginFrame Grid Portal product line
DCV Remote Visualization technology

Other relevant competencies
Distributed Resource Management
Grid Intelligence
Visualization technology integration


NICE Customers and Market Segments

Energy Aerospace & Manufacturing
Addax Petroleum, AECL, Hess, Bayerngas, AIRBUS, Air Products and Chemicals,
BHP Billiton, Beicip, British Gas, Centrica, Procter&Gamble, SelexGalileo, Goodrich
Chevron, Conoco-Phillips, Dong, Dowell, Aerospace, Hamilton Sunstrand, Kimberly
DSC-Libya, ENI/Agip, GazPromNeft, GDF, Clarke, Magellan Aerospace, Nordam,
Logelco, Maersk Oil, Marathon Oil, Nexen, Northrop Grumman, Raytheon, Thales
Novatek, Papuan Oil, Rosneft,
Schlumberger, Shengli Oil, Sonatrach, Automotive & Industrial Equipment
Statoil, Talisman Energy, TNK-BP, TNNC,
Audi, ARRK, Bridgestone, Bosch, Daimler,
TOTAL, WesternGeco, Xinjiang Oil
Delphi, Dow Chemical, FIAT, Ferrari, Honda,
Hyundai, Jaguar-Land Rover, Lear, Magneti
Life Sciences Marelli, McLaren, P+Z, PSA, Tata Steel,
Bayer, Biolab, DEISA project, Swiss Institute Toyota, TRW, Volkswagen
for Bioinformatics, Partners Healthcare,
Pharsight, M.D. Anderson Cancer Center Research & Education
ASSC, Beihang Uni, CCLRC, CILEA, CNR,
Others CNRS/IN2P3, ENEA, Georgia State Uni,
Accent, Samsung SDI, SensorDynamics, INFN, RMSC, Harvard Uni, Messina Uni,
Bank of Italy, Deutsche Bank Huazhong Normal Uni, Yale Uni


The NICE Grand Vision

Internal
Resources
Administrators Partner
& Managers

Supplier

Customer

Multi-Tenancy &
Collaboration services

HPC Centers,
Public Clouds,
…

User


NICE Products & Solutions

NICE EnginFrame

NICE DCV

NICE EnginFrame Views

Solutions for vertical markets
markets:

Life Science
Oil&Gas
Automotive

NICE Products Overview

Project Mgr End Users Developers Developers End Users End Users

Customer 1 (Corporate network) Customer 2 (VPN) ... Customer N

Service A Service B Service C Service X
(HPC) (3D application) (SaaS)
SaaS)
... (workflow)
workflow)

Service Presentation Layer Access Control
(Web, API, Command-line, ...) AD, LDAP, NIS, …

Vertical solutions

Views DataGate
SLB-RE
HPC …
Desktop Cloud
Visualization

Infrastructure model abstraction

GridML ViewsML CloudML Accounting
driver driver driver & Billing

LSF, Lava, SGE, DCV, VNC, RGS, VMWARE, EC2, RDBMS
PBS, MOAB, … TAW, … MOAB, ISF, …


Our architecture
Access Self-Service Offering Resources
Thin viewer

DCV protocol Linux &
Collaborators,
Windows
Support staff
3D
Visualization sessions
Servers

HTTP(S)

HPC schdulers
End Users
HPC jobs

Command-line

Orchestration &
Provisioning
SOAP

Storage

Developers,
Integrators


User Experience in the Technical Cloud
Enterprise Open Grid
Grid Portal

Desktop
Scavenging

Commercial
HPC Cloud
HPC
Clusters


NICE EnginFrame key features

Lightweight Web front-end for HPC systems
Best in class data management
WebServices API and command-line interface
Controlled, user friendly application submission
Supports all mainstream job schedulers
Workload, resource, license monitoring


Batch job / workflow submission

User friendly,
Application-oriented
Job submission

Hide complexity of
Underlying scheduler

Flexible and efficient
Input file management


Application Data Management

Application data
can be organized
into projects

Application data
can be marked as
starred

Metadata can be
associated to
application data


Monitoring

Jobs, Hosts, Queues, Licenses, …


Data transfers & file management

The file manager component allows to seamless navigate and
access server-side files from the web browser


SOA-enabled job submission
WS-I interface
Java / .NET client library and
command line interface
Simplifies integration with
client-side applications
(optimization, workflow, etc.)
for power-users


What is a 3D Cloud?

"Cloud computing is a model for enabling convenient,
on-demand network access to a shared pool of
configurable computing resources (e.g., networks,
servers, storage, applications, and services) that can be
rapidly provisioned and released with minimal
management effort or service provider interaction.” (NIST)

3D Clouds will enable on-demand network access to
interactive 3D applications (like visualization
applications for scientific data, CAD applications, etc.)

Why NICE customers need a 3D Cloud?


A typical scenario in the CAE computing sector

1 - Simulation submission

Win

2 - Data Transfer to visualize results

Server/Data
Center
3 - Collaboration
Needs Data exchange CAE Workstations
• Can run small to medium serial analysis
• Requires high-end GPU for rendering
Win • May represent IT management
challenges (tens to hundreds of seats)


Problems associated to this model

NETWORK
Network overload leading to poor performance and response times all round
COST
Expensive, dedicated workstations (GPU, memory, …) with short lifecycle
IT MANAGEMENT
Support, update and replace tens to tens of thousands of workstations
WORKSTATION SIZING
Workstations have to be sized for the largest expected models
SECURITY
Moving sensitive data around (in/out organization) is always risky
WORKFORCE
Current models do not support a diverse mobile workforce


Benefits of the HPC Cloud model

NETWORK
Network is no more a bottleneck and data loads faster
COST
Centralized & Shared servers are less expensive to buy & manage
IT MANAGEMENT
Support, update and replacement are more efficient & do not affect users
WORKSTATION SIZING
Resources are dynamically sized based on users needs
SECURITY
Sensitive data remain within protected data center with full access control
WORKFORCE
Users can virtually connect & collaborate from anywhere with any client


Enabling Technologies for the 3D Clouds

Technologies providing secure remote network access to
interactive 3D applications leveraging server-side graphic
hardware acceleration (GPUs)

A software stack that allows the end-user to easily launch and
access remote interactive applications and takes care of managing
and load balancing applications and desktop sessions running
within a Visualization Farm


NICE EnginFrame Views

NICE EnginFrame (Portal and Gateway)

Job Scheduler & Distributed Resource Manager

2D & 3D Applications

VirtualGL RealVNC NICE DCV HP RGS NX, RDP, ...

LINUX
WINDOWS
SOLARIS
Computing Infrastructure

Storage Infrastructure


How Does It Work?

The user connects to EnginFrame to
create a new Session
EnginFrame
Job sch. selects the appropriate
node and starts the session
1
Win
2

3

User gets connected to the session
on the selected node
Job Sch.

64GBytes, Linux
4GB, Win/Lin blades
8GB, Windows

Heterogeneous infrastructure: HW, OS, middle-wares
middle-

Benefits for the End Users

Access to Applications as a Service
– Hide infrastructure details (site, platform, etc…)
– Session allocation can be influenced by memory requirements, data
location affinity and other customer-specific parameters
– Intelligent load balancing of sessions, based on the Job Scheduler
• Memory-aware, Memory reservation, Application license-aware
– Fewer or better optimized data transfers
Collaboration, session sharing
– The session owner can generate a “URL” that can be sent by email /
instant message to invite a collegue to join a given session, without
disclosing the user’s password
Easy management of active session
– Create new sessions with user-specific preferences (resolution, etc...)
– List, reconnect and kill existing sessions
Seamless Access to the sessions and Single Sign-On
– Automation of session-level password create/destroy
• E.g. login via NTLM / ActiveDirectory credentials and map to Linux user


Benefits for the Administrators

Increased level of service provided to users
– Sessions are load balanced by the Job Scheduler to match user needs
– Memory reservation and Data locality scheduling
– Reduce help desk calls
– Exposed services can be personalized per user/group/project
Accounting
– Sessions are jobs, so the resource usage accounting by user, group,
project can be collected through any Analytics tool
Monitoring
– The load and usage of the login farm is monitored via EnginFrame
– Node loading conditions, active sessions
– Administrators can control and manage users’ idle or stuck sessions
Support
– Administrators can connect to user’s sessions to provide support
Security
– Easy integration into identity services, SSO, Enterprise portals


NICE DCV (Desktop Cloud Visualization)

Originated in IBM Research in 2004, acquired by NICE in summer
2010

Provides efficient remote access to graphic-intensive, professional
OpenGL applications

A central (graphics-enabled) server is accessed by remote users

Users only need low-end machines with network connectivity to the
server in order to view and interact with remote application

One or more users (collaborators) can simultaneously access the
server

Is the key technology for delivery of real-time 3D graphics but relies
on other software to provide the collaborative environment
(RealVNC Visualization Edition)


NICE DCV – Main Features

Works on Windows and Linux, leveraging high-end NVIDIA GPUs

Support GPU sharing across multiple users, multiple OS
– First product on the market to allow the sharing of physical GPUs
between multiple Windows VMs while maintaining full OpenGL
acceleration and workstation-class performance

Bandwidth optimized balancing of quality Vs. frame rate

High network latency tolerance

Validated and optimized for Technical Computing applications

Perfectly integrated with NICE EnginFrame for session
management


Windows application delivery


Linux application delivery
43


Web-based Thin client access


Built-in collaboration
45


Thank you!


Seminario Paolo Maggi, 24-05-2012

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (18)

En vedette

En vedette (20)

Similaire à Seminario Paolo Maggi, 24-05-2012

Similaire à Seminario Paolo Maggi, 24-05-2012 (20)

Plus de CRS4 Research Center in Sardinia

Plus de CRS4 Research Center in Sardinia (20)

Dernier

Dernier (20)

Seminario Paolo Maggi, 24-05-2012