Meeting Thu Aug 14 at building 28

Participants: Costin Grigoras, Benjamin Gaidioz, Ricardo Rocha, Pablo Saiz, Elisa Lanciotti

Object: verify if the metrics we want to retrieve are available in the Dashboard and Monalisa and decide how to fetch them.

• Data transfer activity.

We want to monitor the total traffic from site A to site B.

For the Dashboard these numbers are available.

For Monalisa: they monitor the xroot traffic from 2 sites. But if they want to monitor the total traffic between two sites, they should add the scheduled transfer via FTS. This is not obvious because if we sum the xroot transfers with the STD transfers then the xrootd traffic is summed twice. A solution would be to take into account also the protocol, in order to be sure not to sum twice the same transfers.

Ricardo observed that in Atlas they distinguish the data transfer in some sub-categories: data transfer of data T0-T1 , data transfer for the analysis jobs, data transfer for the production activity. Currently, they have only 2 categories, but soon (by the end of the month) they will have implemented each category. He says that this distinction is relevant because the different data transfer come from completely independent activities. More: for the T0-T1 data transfer they have a defined target, provided by the experiment (the same is for the other VO, I have to check...), while for the other types of data transfer it is not possible to define a precise target, so they define a status on the basis of the success rate.

If possible it would be useful to have this separation also for the other VOs.

• Job Processing: is it possible to defferentiate the job types?

In principle it is possible on the basis of the user which is owner of the job. But Costin observes that this information is not relevant for the site because there are more reliable ways to ensure that the site is working fine and the nodes are well configured. (see the site tests in Monalisa). This doesn't have high priority.

For the time being, we will keep this distinction for VOs which can provide this information, and for the others no (all the job processing activity will be displayed under 'job processing').

• The site naming convention:

Atlas adopts the GOCDB names, but Alice and LHCb no. So it is necessary to implement a mapping to convert the names of the sites to the GOCDB convention in Monalisa.

• The CPU power normalisation to kspecint

In Monalisa the values of CPU time are normalized in KSI2K. For Atlas and CMS no. They think they should do it, but it is not clear when it will be done. For the time being, we just take into account that their CPU time is not normalized.

• The number of CPUs used

This metric seems not to make sense! First, because it should be, more or less, the same than the number of parallel jobs, second because it is not easy to compute. Apologies, I should not have included in the interesting metrics! let's ignore it.

• How to extract the metrics from the monitoring tools:

Costin proposed via text files in a comma separated format. The others agreed. They will be inserted in a common schema using a collector. Elisa will work on the development of this collector in the Dashboard framework (Pablo has developed something very similar for CMS, so she will ask him for help!).

• Conclusions:

The metrics are defined and Elisa will send today a list with an exact definition of all them.

Not all the metrics are available since now from all the monitoring tools. For example dashboard can't provide the cpu time in ksi2k, but these details can be fixed later.

Any comment about these minutes is welcome! Thank you very much for your collaboration!

-- ElisaLanciotti - 14 Aug 2008