Transcript ppt

Garden of Architectures
CSG Workshop
May 2008
Jim Pepin
CTO
Disruptive change
• Doubling (Moore’s Law or …)
– Transistors
• Multi-core
– Disk capacity
– New mass storage (flash, etc)
– Parallel apps
– Storage mgmt
– Optics based networking
Disruptive Change
• Federated identity
– Large V/O
– Shared research/clinical spaces
• Team science/academics
– Paradigm shift
• CI as a tool for all scholarship
Disruptive Change
• Lack of diversity in computing
architectures
– X64 has ‘won’
• Maybe IBM/Power exists at edges
• Maybe Sun/SPARC at edges
– This creates mono-culture
• Dangerous
– Innovation here in consumer space
• Game boxes/phones drive here
Network Futures
• Optical Bypasses
– Very high speed
• Low friction
• Low jitter
• Facilities based
– GLIF examples
– RONs
– Exchanges
Network Futures
• “Security” is driving researchers away
from us
• Are we the problem?
– Where does ‘security’ belong?
• How do we do VOs with two port
internet?
• Will we see our networks become
‘campus phone switch’ of the 2010s
•
Data futures
•
•
•
•
•
•
Massive storage (really really big)
Object oriented (in some cases)
Preservation
Provenance
Distributed
Blur between data bases/file systems
– Meta data
New Operating Environments
• Operating systems in network
– Grids
• ID management
– But done poorly from integration view
• How to build petascale single systems
– Scaling applications is biggest problem
• Training
• “Cargo Cult” systems and applications
New Operating Environments
• 100s of TF at campus (but how to use it
and build it on campus)
– Tied into national petascale systems
– All the problems on terragrid and VOs on
steroids.
• Network security friction points
• Identity management
• Non-homogenous operating environments
Computation
• Massively parallel
– Many cores (doubling every 2-3 yrs)
• Commodity parts
– Massive collections of nodes with high
speed interconnect
• Heat and power density
• Optical on chip technology
– Legacy code scales performs poorly (or
worse)
Viz/remote access
• SHDTV like quality (4k)
– Enables true telemedicine and robotic
surgery
– Massive storage ties to this
– Optiputer project is example (CALIT2)
– Colab spaces with true haptic and visual
presence.
• Social sites are simple prototypes
• Large screen applications and tele-presence
Versus
• Old Code
– Much based on 360/VAX/Name it
• Gaussian poster child
– Vector optimized
• Static IT models
– Network defenders in IT hurt researchers
– Researchers don’t play with others well
– Condo model evolving
Versus
• Thinking this is just for science/engineer
– Large data
– Interactive applications
• Social Science apps
– Education outcomes at Clemson
• Large data, statistics on huge scale
– Shoah Foundation at USC
• Massive data, networks, VO
Vision/Sales Pitch
• Access to various kinds of resources
– Parallel high performance
• Can be in condo (depends on politics)
– Flexible node configurations
– Large storage of various flavors
– Viz
– Leading edge networks
“Clusters”
• Large collection of multi-core
– High performance interconnect
• What makes cluster not just a bunch of nodes
– Access to large data storage at parallel
speeds
• Lustre
• SAM/QFS
• PVFS
– Ability to put in large memory nodes
“Clusters”
– Magic chips
• GPUs, FPGAs etc
• Botique today but gains can be enormous
– Relation to desktops/local systems
– How to integrate into national systems
• Identity/security/networking
– Viz clusters
• Render agents
• Large scale, friction free networking
Storage Farms
• Diverse data models
– Large streams (easy to do)
– Large number of small files (hard to do)
– Integrate mandates (security, preservation)
– Blur between institution data and
personal/research
– Storage spans external, campus,
departmental,local
– Speed of light matters
Meaning of Life
• Much closer relations needed to central
IT
– Networks/identity mgmt/security/policy
– But not just ‘at scale’
• How to use the disruptive technologies
– Core,GPUs,Cell,FPGA,Flash,optical
networks
– Disruptive software/services as well
Meaning of Life
• Build ecosystem of services
– Some central, some local, some external
– Not just computing, networks and storage
– Our community has “gone global”
•
•
•
•
The campus is not a castle.
Earlier example of 8 social science faculty
We have thousands of communities
Can’t be one size fits all