Title: Developing Grid and WebBased Applications
1Developing Grid and Web-Based Applications
Opportunities Challenges
Shannon Whitmore San Diego Supercomputer
Center swhitmor_at_sdsc.edu
2Introduction Seti_at_Home
- Basic ideas
- distribute small work units to many compute
resources - utilize idle compute cycles
- build redundancy into the system
- Simple client application
- available to the public
- ported to a variety of platforms
3Intro Seti_at_Home--Results
- Over a 12 month period
- 1.7e21 floating point operations
- 221 million recorded work units
- Average throughput 27.36 teraflops
- Participants in 200 countries
- Client running on 140 platforms
- 1.7 million CPU hours total
- 1.5 million on x86 platform
4Intro Grid Motivation
- Extend computing power
- Faster time to results
- More sophisticated analysis
- Lower computing costs
- Maximize use of existing resources
- Separate data acquisition from computation
- Decrease dependence on one system
- Facilitate collaboration
5Why isnt grid computing as easy as SETI appears
to be?
6Interactive
- How many of you have opinions on the grid?
- How many of you have run applications on the
grid? - If you have, then you know that running
applications is not always easy
7People
- Resource site staff
- Middleware developers
- Grid applications developers
- Users
8Grid Layers
Hardware Software (Compute, Memory, Disk,
Application, Network)
Resources
Virtual Operating System (Authentication
Security, Access, etc)
Middleware
Applications
High Level Software
Grid Consumer
9Grid Environments
- Geographically dispersed resources
- Different rules for access
- establishing accounts
- tracking usage
- managing security
- Heterogeneous systems
- hardware
- operating system
- middleware
- configuration
10Grid Layers
Hardware Software (Compute, Memory, Disk,
Application, Network)
Resources
Virtual Operating System (Authentication
Security, Access, etc)
Middleware
Applications
High Level Software
Grid Consumer
11Middleware
- Rapidly changing standards
- Heterogeneous systems interoperability
- Resources come and go
- Consists of large code base
- Not robust or hardened
- Incomplete
- Developed by different organizations
12Why the Complexity (contd)
- Many components exist
- What do they do?
- Which ones to use?
Development Tools/Libraries GridPort
(Pending) MPICH-G2 LaPACK for Clusters
Distributed Data Management Datacutter Storage
Resource Broker Globus (GridFTP/GASS)
Job Management Globus (GRAM) APST Condor-G
Monitoring/Prediction Network Weather
Service Ganglia Globus (MDS)
Authentication GSI-OpenSSH MyProxy Kx.509
Configuration Management GridConfig
13Grid Layers
Hardware Software (Compute, Memory, Disk,
Application, Network)
Resources
Virtual Operating System (Authentication
Security, Access, etc)
Middleware
Applications
High Level Software
Grid Consumer
14Grid Applications
- Middleware does not guarantee a minimum
environment - Environment variables
- Existence/Location of software
- Grid services may not be available when needed
sometimes software is down - Applications are difficult to debug
- Pinpointing problems across distributed resources
- Some software does not expose debugging
information
15Grid Layers
Hardware Software (Compute, Memory, Disk,
Application, Network)
Resources
Virtual Operating System (Authentication
Security, Access, etc)
Middleware
Applications
High Level Software
Grid Consumer
16Grid Consumers
- Access to a grid is not always easy
- Obtain an accounts at all sites
- Obtain certificate
- Certificate propagation involves multiple steps
- Grid applications are complex
- Setup/configuration steps
- Complexity from lower levels is not hidden
- Most users need to be grid experts
17Challenges Summary
- Its a complex system in its infancy
- Middleware wont immediately stabilize
- Standards will take a while to settle
- Tremendous potential
- Burden at this time is on the grid application
developer
18People
- Resource site staff
- Middleware developers
- Grid applications developers
- Users
19What can be done?
20Opportunities
- Hide complexities of lower layers
- Users should only have to know enough
- to run their application
- Make software fault tolerant
- Develop reliable and easy to use interfaces
- Provide documentation and user support
21Opportunities (Contd)
- Automate as much as possible
- Grid accounts and certificates
- Develop Web Portals/Services
- Single point of access to the grid
- Simplifies workflows
- Spend time developing glue code
- middleware integration
- missing functionality
- error checking
22Opportunities (contd)
- Work on narrower problems
- Make a smaller number of tools work well
- Develop relationships
- With users, resource sites, and middleware
developers - Get more involved in standardization process
23Case Study
24Scalable Visualization Portal
An application for rendering frames of raw
volumes across heterogeneous grid resources
Your Job
SRB
Volume Archive
25VisPortal
- Hardware
- NPACI Grid
- Horizon, Longhorn, Morpheus
- Teragrid
- SDSC
- Software
- NPACKage
- APST, GridPort, SRB, Globus GSI-OpenSSH
- Scalable Visualization Toolkit
- Vista (renderer)
26NPACI Grid
Collectively, the NPACI Grid provides 1766 CPUs
and 4190 peak GFlops of computing power.
Longhorn
Hypnos
Morpheus
Blue Horizon
27Grid middleware
28Under the Hood
Grid Resources
SRB Server
Job Submission Host
Portal
Raw volumes Rendered images
Longhorn Blue Horizon Morpheus Hypnos
Teragrid APST SRB Client Vista
WebServer VisPort interface GridPort Globus GSI
Open SSH
Griddle APST Globus GSI Open SSH
User Workstation
Web Browser SRB client
29Job Execution
30VisPortal Live
Demo
31Acknowledgments
32- Additional Information
- http//npackage.npaci.edu
- http//npacigrid.npaci.edu
- http//vis.sdsc.edu
- http//vistools.npaci.edu
- http//visservices.npaci.edu
swhitmor_at_sdsc.edu
grid-consult_at_npaci.edu
33SCEC Visualization
- Rupture dynamics simulation
- Rupture event between 2 faults at the
hypocenter, located underground - Velocity wave front shown at the north-south
strike-slip fault - 3000 time steps
- every 5th step saved
- 600 frames
- 1 GB sized files