WLCG Accounting Task Force Meeting
Attended:
Maarten, Dimitrios, Jarka, Gavin, Alessandro, Davide, Greg, Pepe, Stefan, Concenzio, Julia
We had time only for discussion of the first topic "RRB reports"
ALICE
Numbers for tapes look suspicious.
Julia mentioned about the process of enabling tape accounting numbers in the storage space accounting system. Currently 5 sites are there.
For internal disk accounting, ALICE monitoring system is used. It is enabled for xrootd and EOS. For dCache and DPM data is missing, though for ALICE this is a small fraction of sites
Dimitrios commented that ALICE storage space accounting will be soon visible in the Storage Space accounting system
According to ALICE, CPU accounting in the EGI portal is OK. CERN is fine as well.
ATLAS
HLT and CAF is not in EGI, ATLAS internal accounting is used for it.
EGI is mostly used for T1 and T2, though there are some inconsistencies discovered.
Julia mentioned EGI accounting validation system and suggested ATLAS sites to check it on regular basis.
Not clear how/who can enforce checking of the correctness of the accounting numbers published in the accounting reports. Should be a responsibility of the sites. Experiments do not have effort to follow it up. The latest example is the problem with CNAF accounting in September & October. Though the problem is visible in the validation system here,
it was noticed and fixed only when Davide prepared the RRB reports.
There are metrics like number of produced events which are included in the reports, which certainly is accounted only in the experiment-specific sources. Same is true for data popularity metrics.
HPC and commercial cloud usage is currently also coming from the experiment sources. This usage can be imported into EGI portal in case benchmarking of such resources is available.
For storage space accounting, internal ATLAS accounting is used.
Some conclusions for ATLAS:
A lot of various tools need to be used. Some of those can be only ATLAS-specific.
There is a concern regarding the quality of the plots provided by new MONIT-based monitoring system. In order to get good quality plot, people need to import data in csv in some system of their preference and to recnstruct the plot on their own. May be we can think how to facilitate this task.
Known problem of CERN accounting is to be followed up.
CMS
Same as ATLAS, a lot of systems, among those many are experiment-specific, like CMS Dashboard. Some metrics like split of production/analysis can come only from the experiment-specific systems.
Complicated cases like data popularity and number of disk accesses for certain data.
Julia suggested to coordinate between experiments to agree on set of metrics to be included in the reports, since some of them with time become not so useful, while new ones show up. LHCb temped to agree with this coordination, while ATLAS representatives think that it should be rather RRB who decides.
Problem with xrootd monitoring and correspondingly accounting
Similarly to ATLAS, many plots are created manually out of data retrieved from various monitoring systems
LHCb
Same as other experiments. Most of sources of information are experiment-specific, though EGI accounting portal is used and considered to be pretty reliable
-----
Did not have time to discuss CERN-ATLAS accounting inconsistency. Decided to organize a meeting during the WLCG workshop in Naples.
Next meeting is in approximately one month. Julia will send an announcement.