1. The UberCloud
Paving the way to
High Performance Computing as a Service
International Conference on Utility and Cloud Computing
December 9 – 12, Dresden, Germany
Wolfgang Gentzsch and Burak Yenier
2. What is High Performance Computing ?
Modeling the World on the computer …
t U U U 0
F ma
i t H
E mc 2
… in order to allow for Virtual Product Development …
2
3. What is High Performance Computing ?
… or for scientific insights
Combustion
Environment
Cosmology
Climate
3
4. HPC is needed to stay competitive
The digital manufacturing engineer has
several options to use High Performance
Computing (HPC):
HPC on the Desktop: over 90% of engineers
HPC on the Server: about 5% of engineers
HPC as a Service: in the Cloud; less than 1%
5. Workstations: limited capacity
Low-end workstations and PCs are important for
daily design and development work, but
50+ % of users are dissatisfied with their computing
capacity*
Too slow, e.g. jobs run over night or a whole week
Too small, detailed geometry and physics don’t fit into
memory
Number of jobs are limited which affects the quality of
the final result
*
Source: http://www.compete.org/
6. Servers: expensive and complex
For SMEs buying and using large scale HPC systems is
expensive and complex …
7. HPC as a Service: benefits & challenges
HPC as a Service (in the Cloud) offers flexibility, business
agility, scaling up and down, pay-per-use, OPEX instead
CAPEX, but
It’s a new business and working paradigm
Security, privacy, trust in service provider
Intellectual property
Software Licensing
Heavy data transfers
9. The UberCloud HPC Experiment
An open voluntary collaborative community
Objective:
Making HPC as a Service available, for everybody, on demand
How?
For SMEs and their engineering applications
to explore the end-to-end process
of using remote computing resources,
as a service, on demand, at your finger tip,
and learning how to resolve the roadblocks.
10. The End-User’s Benefits
Free HPC Experiment, on-demand access to hardware, software,
and expertise, with a one stop resource access experience
No hunting for resources in the complex emerging cloud market
Professional match-making of end-users with service providers
Perfectly tuned end-to-end, step-by-step process to HPC Cloud
Lowering barriers & risks for frictionless entry into HPC Cloud
Crowd sourcing: End-Users are building relationships with other
community members who actively contribute to improvements
11. How does the Experiment work?
End-User joins the experiment
Software Vendor joins
We select a Team Expert
We suggest a Resource Provider
Team is ready to go
… 22 steps on Basecamp’s virtual team office
Finally, writing the Case Study
12. Where are we with the experiment
Started August 2012: today 700+ participating organizations
and individuals
Participants are from 66 countries
Round 4 started September 1: already 42 new teams
124 teams have been formed in Rounds 1-4
Registration at:
www.hpcexperiment.com
www.cfdexperiment.com
www.compbioexperiment.com
www.bigdataexperiment.com
13. UberCloud community website
With social network and crowdsourcing features
Forums, Q&A, discussions, newsletters,…
Feature stories: HPCwire, Desktop Engineering, Bio-IT…
UberCloud University with free and paid lectures
UberCloud HPC Experiment free trial service
UberCloud Exhibit services directory
List of currently ongoing team projects, their status, and
organizations involved
List of upcoming conferences with ‘meet me there’ button
16. Finally: The UberCloud Marketplace
Crowdsourcing and social networking: provide the
community web platform to discover and try HPCaaS,
address pain points, facilitate adoption, harness collective
intelligence
Marketplace for computing-related services: 20+ million
engineers, scientists, and their service providers to list,
discover, try, and sell/purchase
IaaS: HPC centers and public cloud providers
SaaS: open source and commercial technical software
Expertise: specialized HPC, software, technical know-how
17. Cloud Computing Reality Check
UberCloud Compendium
sponsored by Intel
25 selected use-cases from
60 teams in Rounds 1 & 2
Google:
“ubercloud compendium”
17
18. Team 2: Simulating new probe design
for a medical device
Front End + 2 GPU Solvers In Action
19. Team 2: Team members
HPC Expert:
Electromagnetic Simulation
End User: Anonymous
20. Team 8: Multiphase flows within the
cement and mineral industry
Drying moist particles with hot gas
21. Team 8: Team members
Ingo Seipp
and Team
End-User
CFX
22. Team 26: Simulating Stent
Deployment
Using SIMULIA’s Abaqus/Standard and Remote Viz
Software from NICE to run CAE on SGI Cyclone™
Assessment of a fictitious balloon expandable stent design
deployment, physiological pulsatile loading
and both axial and radial compression.
23. Team 26- Team members
End User: Anonymous: Global Designer and Manufacturer of Sterile
Medical Products
CAE Software Provider: Matt Dunbar, Chief Architect, SIMULIA
Remote Viz Software: NICE Desktop Cloud Visualization (DCV)
HPC/CAE Expert: Scott Shaw, Senior Applications Engineer, SGI
Resource Provider: SGI Cyclone. Tony DeVarco, Senior Manager for
Strategic Partners and Cloud Computing at SGI. Eugene Kremenetsky,
Systems Engineering Technical Lead at SGI
Team Mentor: Gregory Shirin from the HPC Experiment team
24. Challenges and Solutions
Information security, privacy: protecting the users intellectual
property, guarding raw data, processing models, resulting
information: document security requirements, and select the
right provider
Internet too slow for heavy data transfer: don’t ship every
result, just the important ones; use remote visualization; if
necessary, fedex the data over-night via USB hard drive
Incompatible software licensing models: ISVs have to develop
compatible on-demand software licensing models
25. Challenges and Solutions, cont’d
Reliability & availability of resource providers: seeking info on
reliability and availability of each vendor before partnering
Lack of easy registration and administration: originally HPC
resources are not designed for the masses: use automated rules
based instant decision making capabilities
Costs: pay-per-use billing can result in unpredictable cost,
project can easily run out of budget: use automated, policy
driven monitoring of usage and billing
High expectations, disappointing results: we are in transition
to the cloud computing paradigm: Set goals that are
incrementally better than your current capabilities.
26. Why join?
HPC as a Service is the next big thing, benefits are obvious
HPC is complex; together it is easier to tackle the complexity
Low entry barrier to HPC as a Service through an experiment
Learning by doing; experimenting with no risk; no failure
Becoming an active part of this growing community
Exploring the end-to-end process and learning how this fits
into your research and/or business direction in the near future