High Energy Physics @ FermiLab

Two physics detectors (5 stories tall each) to understand smallest scale of matter

Each experiment has ~500 people doing science

Each experiment handles millions of particle collisions per second - HUGE amount of data!

Data volume and analysis rates

Total 4.3 petabytes

Analysis >1 petabyte/month

The data challenge

Use resources at participating institutions Ship and analyze data around the world !

Get data fast, processes it, and then immediately store the results back at Fermilab

Strategy

Common business model for data cataloguing, tracking, and mining.

Streamlined support of the underlying machineryShared expertise solves issues at the user level.

Follow grid standards, use grid middleware and shared resources (OSG and LCG grids).

Contribute to grid projectsOSG resource selection, SRM, security…

How we handle data

Sequential access via Metadata ( SAM )data storage, directly from the detector or

from remote data processing facilities data cataloguing, miningdistributed resources management to

optimize usage and data throughput, and enforce the policies of the experiments.

Use of variety storage service providersdCache, enstore, HPSS, SRM, in house disk

resources

Success story: D0 refixing

Problem. Correct processing mistake in 6 weeks 85 Tb and 4Million hours of 1Ghz CPU time

Plenty of network but no free CPU to do the job at Fermilab

Solution. Involve CPU resources elsewhere. Ship detector data directly to the analysis Cache re-usable data near computing sites

OSG grid 20Tb

LCG grid 5Tb

Westgrid 10Tb Fermi 40Tb IN2P3 10Tb

Conclusion

Store a petabyte a yearProcess a petabyte a monthComputing that meets growing demands

of the HEP experimentsone step ahead of the physics needs