CPU_SCALING_FOR_ACCOUNTING

[kreczko]

This is a grid-specific feature but might be interesting for other accounting purposes as well.

A few words on CPU scaling for accounting:
The grid uses HEPSPEC06 to benchmark a machine. This benchmark and related ones (SPECINT2000, SPECFP2000 == 250 * HEPSPEC06) are used to evaluate resources used by a job. Historically all nodes of the same type/HEPSPEC06 value would be grouped into separate clusters. However, it is possible to run different nodes within the same cluster when rescaling the HEPSPEC06 value to the published base line.

Example: A cluster contains two types of machines, type 1 with HEPSPEC06 = 20 per core and type 2 with HEPSPEC06 = 10 per core. The published value is the average, 15 HEPSPEC06. For accounting purposes the used CPU resources for type 1 nodes need to be scaled up (factor 20/15) and scaled down for type 2 nodes (10/15). The RAL implementation of this is shown in ARC+HTCondor Accounting & Example ARC CE+HTCondor build and a general example is given in Publishing tutorial.

For the completion of this ticket I suggest:

• add CPU_SCALING_FOR_ACCOUNTING as facter & HTCondor ClassAd:

CPU_SCALING_FOR_ACCOUNTING = 1.2
STARTD_ATTRS = $(STARTD_ATTRS) CPU_SCALING_FOR_ACCOUNTING

• add HEPSPEC06, SPECFP2000, SPECINT2000 to facter & HTCondor ClassAd