Upload
us-ignite
View
53
Download
0
Embed Size (px)
Citation preview
1 Beyond Today’s Internet • March 25, 2015
Beyond Today’s Internet Experiencing a Smart Future
Science DMZ as a Service: Creating Science Super-Facilities with GENI
Inder Monga, Chin Guok, Eric Pouyoul: ESnet Ilya Baldin, Paul Ruth: RENCI Simon Patton, Craig E. Tull: Berkeley Lab
2 Beyond Today’s Internet • March 25, 2015
Inder Monga Ilya Baldine
Simon Pa2on Eric Pouyoul
Chin Guok Craig Tull
3 Beyond Today’s Internet • March 25, 2015
DOE Facili*es
4 Beyond Today’s Internet • March 25, 2015
Emerging Trend: Super Facilities, Coupled by Networks
Experimental facili*es are being transformed by new detectors, advanced mathema*cs, robo*cs, automa*on, advanced networks.
5 Beyond Today’s Internet • March 25, 2015
6 Beyond Today’s Internet • March 25, 2015
• Complexity of scien*fic discovery increasing
• Data volumes are increasing > Moore’s Law
• Fewer large facili*es, but global scien*fic popula*on
Automated coupling of compute, storage with networks cri*cal to increasing science produc*vity
7 Beyond Today’s Internet • March 25, 2015 Slide from Craig E Tull, LBL
8 Beyond Today’s Internet • March 25, 2015
Synchrotrons shed new light onto Sciences Very diverse Science Impacts
9 Beyond Today’s Internet • March 25, 2015
Super-Facility Illustrative Data Flow
Data collection
Transfer to NERSC
FFT + mask
data from experiment
Analysis and modeling on NERSC supercomputers: HipGISAXS simulation HipRMC fitting
FFT
Compare
start with random system move parCcle random
Autotuning
On-the-fly calibration, processing
Combining:GIXSGUI, dpdak + …
Real-time access via web portal
10 Beyond Today’s Internet • March 25, 2015
Global data movement infrastructure has challenging end-to-end requirements
Physical pipe that leaks water at rate of .0046% by volume. è è
Network ‘pipe’ that drops packets at rate of .0046%. è è
Result 100% of data transferred, slowly, at <<5% opCmal speed.
Result 99.9954% of water transferred, at “line rate.”
essenCally fixed
determined by speed of light
Through careful engineering, we can minimize packet loss.
AssumpCons: 10Gbps TCP flow, 80ms RTT. See Eli Dart, Lauren Rotman, Brian Tierney, Mary Hester, and Jason Zurawski. The Science DMZ: A Network Design Pa`ern for Data-‐
Intensive Science. In Proceedings of the IEEE/ACM Annual SuperCompuBng Conference (SC13), Denver CO, 2013.
11 Beyond Today’s Internet • March 25, 2015
End-‐to-‐end architecture is cri*cal Science DMZ has three key components, all required: • “FricCon free” network path
– Highly capable network devices (wire-‐speed, deep queues) – Virtual circuit connecCvity opCon – Security policy and enforcement specific to science workflows – Located at or near site perimeter if possible
• Dedicated, high-‐performance Data Transfer Nodes (DTNs) – Hardware, operaCng system, libraries all opCmized for transfer – Includes opCmized data transfer tools such as Globus Online and GridFTP
• Performance measurement/test node – perfSONAR
• Note: General Atomics Science DMZ first in world (we think) where IPv6 was chosen for performance reasons.
Details at h`p://fasterdata.es.net/science-‐dmz/
© 2013 Wikipedia
12 Beyond Today’s Internet • March 25, 2015
Representative Science DMZ
13 Beyond Today’s Internet • March 25, 2015
• Current implementations deploy dedicated DTNs in the Science DMZ
– Manual configuration and tuning
– Sharing is scheduled, many times manually
Dedicated vs. Virtual resources
14 Beyond Today’s Internet • March 25, 2015
Dedicated vs. Virtual resources • GENI provides a distributed solware-‐defined infrastructure
(SDI) – Compute + Storage + Network
15 Beyond Today’s Internet • March 25, 2015
• GENI provides a distributed software-defined infrastructure (SDI)
– Compute + Storage + Network
• GENI racks may be deployed on-campus or in provider networks close to the campus
• ‘Science DMZ as a service’
– Applications can provision a virtual ‘Science DMZ’ as and when needed
Dedicated vs. Virtual resources
Programmable infrastructure to enable end-‐users to create dynamic ‘fric*on-‐free’ infrastructures without advanced knowledge/training
16 Beyond Today’s Internet • March 25, 2015
Today’s Demonstration: Real-time data processing and vis. workflow
h`p://portal.nersc.gov/project/als/sc14/
Data from ALS Experiment
SPADE instance @ Server at Argonne
ExoGENI SPADE VM @ Starlight, Chicago
ESnet
ExoGENI SPADE VM @ Oakland, California
Compute Cluster NERSC, LBL
AL2S, ESnet
• WAN-‐op*mized data transfer nodes and a network slice created programma*cally (Science DMZ as a service)
• Applica*on workflow instan*ated to stage data at the GENI rack on Science DMZ slice
• Data is moved op*mally across the WAN1
1 Earlier work, like Phoebus, have instanCated the value of this approach
17 Beyond Today’s Internet • March 25, 2015
Microtomography of High Temperature Materials under stress
Set collected by materials scienCst Rob Ritchie, LBNL/UCB
18 Beyond Today’s Internet • March 25, 2015
Summary and Future • Many science applicaCons are now building coupled workflows across the
WAN to Ce together specialized science instruments for ‘big data’ discovery
• Programmability extends to the enCre ‘Solware-‐defined infrastructure’ stack as illustrated by GENI racks
• WAN ‘Infrastructure as a Service’ soluCons will become pervasive, as programmability and virtualizaCon moves outside the data center domain.
• We think of the ‘network’ as an ‘instrument’ that enables scienCfic discovery
• Should there be a disCncCon between a network and a data center?
• What will the next generaCon WAN look like?
19 Beyond Today’s Internet • March 25, 2015
THANK YOU
Acknowledgements: DOE ASCR Funding and Support, Jason Lee, Brent Draney and NERSC networking team, Dula Parkinson (ALS), Linda Winkler and Argonne Networking Team (ANL) [email protected], [email protected], [email protected]