PDD - Computing Infrastructure
By Section | Whole Paper | PDF (233 pages, ~6.62mb)
9 Data Analysis- 9.1 Introduction
-
9.2 Analysis Infrastructure
- 9.2.1 Calibration Analysis and Data Quality Working Group
- 9.2.2 Simulation Working Group
- 9.2.3 Reconstruction Working Group
- 9.3 Computing Infrastructure
- 9.4 Physics Analysis
-
9.5 Internal Review Procedure
- 9.5.1 Introduction
- 9.5.2 Procedure
- 9.6 Prerequisites and Schedule
9.3 Computing Infrastructure
Computing resources are needed for data distribution, data analysis and simulation. Two copies of all filtered data reside on disk in two post-filter clusters, one in the US and the other in Europe. This data amounts to roughly 2 TB/yr. In addition to the servers needed to control these disk caches, permitting ready access to all members of the collaboration, each cache also has an associated ∼50 GHz of CPU to perform additional filtering and/or reconstruction tasks. This cluster is built up in stages to match the rate at which data is acquired and to minimize the total expense.
Each FTE in data analysis is provided with a small "local" computing cluster (on the order of four machines with ample disk space and some peripherals, replaced every three years) for use in analyzing data. These clusters reside at each user's home institution, and are probably ganged together when multiple FTEs in data analysis are at the same institution.
Acquisition of a large cluster for simulating IceCube and IceTop data is also planned. The cluster also filters and caches this data. The nominal goal is to keep pace with the actual data rate. Assuming roughly 0.5 s/event/GHz-CPU for generation and simulation, this translates to a need for 800 GHz CPU power for the full detector. As with the post-filter cluster, this cluster is also built up in stages.
