Dmytro Karpenko

Overview

This page complements "Atlas Grid Workload on NDGF resources: analysis and modeling" article (read here). The purpose of the article was to analyze that part of the workload of ATLAS experiment at CERN that is being processed on NDGF resources. The goals of the analysis were to describe the features of one of the most famous and demanding modern scientific workflows and build its model, so the workload can be recreated on smaller scale for the purpose of simulation. Since this page is just a collection of appendices to the article, we will mostly not repeat numbers, tables, figures and text that is already there; we focus on publishinig here the data that did not make it to the article itself.

The collected workload spans from 1 February 2011 to 31 July 2011.

Jobs

• Total: 2,129,612
• Succeeded: 1,940,260
• Successful jobs that registered transfer: 1,674,672

When we say "succeeded" we mean that the job was successful from grid's point of view. I.e., finished normally in LRMS, had zero exit code, was logged as "FINISHED" by the grid middleware. It does not mean that the job was successful from ATLAS's point of view. It is possible, that the job could not by some reason perform the needed physical analysis and instead of producing output data it produced error report, that was uploaded to the storage instead of expected output. Such jobs would usually be resubmitted again. But the job does not crash, so for a local grid administrator it finishes absolutely correctly and such jobs constitute a part of normal grid workflow.

Transfers

• File downloads: 11,069,724
• Unique input files: 2,837,488

Some averages

• Average file size for input files: 338 MB
• Average file size for output files: 100 MB
• Average file size for all files: 205 MB
• Average number of input files per job: 6.6
• Average number of output files per job: 2.2
• Average total input size per job: 1.7 GB
• Average total output size per job: 218 MB