ATLAS First Run Computing Requirements

The following are the resource expectations for the first run in 2007/2008. The model and numbers are essentially those of the document `Principles of Cost Sharing for the ATLAS Offline Computing Resources', http://www.quark.lu.se/~torsten/Computing-resources.pdf. However, there has been some rebalancing between tape and disk usage. The most obvious changes are: the reduction to one full copy of the raw data stored at CERN, and the assumed trigger rate and event sizes have been reduced. Partial copies of the full ESD are stored at each Tier-1, and it is assumed that there is inter-access between the Tier-1 facilities. In this way, disk resident multiple copies of each event are always available at ESD level to the Tier-1 cloud.

We assume 1E7 seconds of running spread over the second half of 2007 and the first of 2008. The following `per event' numbers are assumed. They are based on the optimised FORTRAN code. At present, the ATLAS Object Oriented code does not run this quickly. Further, until the POOL integration exercise is complete, we will not have a better idea of the true event size at the various processing stages. 

The following gives the `external' inputs such as the data rate from the detector. The trigger rate has changed since the `Hoffman' LHC computing review.

For the Tier-1 and Tier-2 resources, a major component is the user analysis activity. The numbers for the Tier-1 analysis activity is taken from the LHC computing review and the Principles of Cost Sharing document. However, these need revision, and we are in the process of capturing the physics resource usage patterns for our recent Physics Workshop as a guide. The Tier-2 facilities will be used even more intensively for the physics analysis. We have therefore assumed somewhat larger resource allocations per physicist in the Tier-2s, which are not an unreasonable extrapolation of the University growth in University resources per physicist in recent years.

Tier-0

It is assumed that the Tier-0 will hold the raw data on tape and the calibration data on disk. It will also have a full copy of the ESD on disk and the previous version on automatic tape. The Tier-0 CPU will perform the first pass reconstruction and one reprocessing. 

CERN Tier-1

The CERN `Tier-1' will hold 1/3 of the current ESD on disk and 1/6 of the previous ESD on disk and on tape, which will also be true for the external Tier-1 facilities. This saves tape storage, as there will always be multiple copies of a given ESD event in the current processing on disk in the Tier-0 and Tier-1s.  The detailed assumptions are as follows:

The total required CERN resources are given at the end. These are the inputs to the LCG cost-to-completion exercise.

External Tier-1

The Tier-1 resources are assumed to provide both the simulation and be used for analysis. As yet we do not plan to have copies of the raw data in the Tier-1s, other than small samples. It is envisioned that there will be approximately 6 external Tier-1 facilities. The resource expectations are as follows:

Tier-2

We picture four Tier-2 facilities for each Tier-1 on average, each with about 50 users. Again, these are used for simulation and analysis. The expected resource requirements are as follows:

Summary

The total resource requirements are summarised below:

This page was constructed for the Computing Model Group by Roger Jones (Roger.Jones@cern.ch) on 16th August 2003