Public Abstract

89243025SSC000137: Preparing for the Exascale of the CMS experiment

Award Status: Active

Institution: NAVY, UNITED STATES DEPARTMENT OF T, Annapolis, MD
UEI: C3FJGE4ZRPD5
DUNS: 612170217

Most Recent Award Date: 06/05/2025
Number of Support Periods: 1
PM: Love, Jeremy

Current Budget Period: 03/01/2025 - 02/28/2027
Current Project Period: 03/01/2025 - 02/28/2027
PI: Hall, Allison

Supplement Budget Period: N/A

Public Abstract

Preparing for the Exascale of the CMS experiment Through the support of this grant, the PI, Dr. Allison Hall, will build research capacity at USNA to develop solutions to critical computing challenges faced by the Compact Muon Solenoid (CMS) experiment in the Exascale era through collaborations with scientists and computing professionals at Fermi National Accelerator Laboratory (FNAL), including co-investigators Dr. Bo Jayatilaka and Dr. Nick Smith. The High Luminosity LHC (HL-LHC) upgrade of the LHC will drastically increase the collision rate and will enable CMS to search for even rarer physics processes. This huge influx of data, however, presents considerable computing challenges in how to efficiently store and analyze exabyte-scale datasets. The primary objective of the proposed computing research is making a credible plan for how to transition the current CMS computing model (including both central offline production and user analysis) to one that takes advantage of the benefits of object stores. In particular, the Fermilab Tier 1 computing center is mission critical for the success of the entire experiment and is the largest dedicated CMS computing resource outside of CERN. Currently, CMS uses a tiered data storage approach, with the RAW detector output stored on tape and end-user analysis formats stored on disk. At each data tier, only a subset of the total information is kept. If an analysis needs data from earlier tiers, the entire dataset needs to be staged to disk and reprocessed. This is an issue especially for analyses (such as those the PI has been leading looking for long-lived particle decays) that rely on low-level detector information for optimal results. Fermilab has been exploring a novel solution to this problem using object stores. Object stores, compared to the current file-based organization of CMS data, allow for highly granular data access. The feasibility of this approach has been demonstrated, and the PI and postdoc supported by this grant will work closely with Fermilab to make a road map for implementation at the Fermilab Tier 1, including a proof-of-concept test using cloud computing resources.

Preparing for the Exascale of the CMS experiment

Through the support of this grant, the PI, Dr. Allison Hall, will build research capacity at USNA to develop solutions to critical computing challenges faced by the Compact Muon Solenoid (CMS) experiment in the Exascale era through collaborations with scientists and computing professionals at Fermi National Accelerator Laboratory (FNAL), including co-investigators Dr. Bo Jayatilaka and Dr. Nick Smith.

The High Luminosity LHC (HL-LHC) upgrade of the LHC will drastically increase the collision rate and will enable CMS to search for even rarer physics processes. This huge influx of data, however, presents considerable computing challenges in how to efficiently store and analyze exabyte-scale datasets. The primary objective of the proposed computing research is making a credible plan for how to transition the current CMS computing model (including both central offline production and user analysis) to one that takes advantage of the benefits of object stores. In particular, the Fermilab Tier 1 computing center is mission critical for the success of the entire experiment and is the largest dedicated CMS computing resource outside of CERN. Currently, CMS uses a tiered data storage approach, with the RAW detector output stored on tape and end-user analysis formats stored on disk. At each data tier, only a subset of the total information is kept. If an analysis needs data from earlier tiers, the entire dataset needs to be staged to disk and reprocessed. This is an issue especially for analyses (such as those the PI has been leading looking for long-lived particle decays) that rely on low-level detector information for optimal results. Fermilab has been exploring a novel solution to this problem using object stores. Object stores, compared to the current file-based organization of CMS data, allow for highly granular data access. The feasibility of this approach has been demonstrated, and the PI and postdoc supported by this grant will work closely with Fermilab to make a road map for implementation at the Fermilab Tier 1, including a proof-of-concept test using cloud computing resources.