Choose timezone
Your profile timezone:
- Current goal is the fulfilment of the PoC milestone.
- No progress with Tier-0 jobs sent via HammerCloud. Very likely an obscure issue with the job definition.
- Previous jobs with an obsolete software version would run to the first event and then throw a code exception. But they would get through the initialisation phase, which was the first issue to solve.
- Real ATLAS prod jobs with different workflows (both single and multi-core) have been allowed in the system for extended periods and have proved to perform reliably.
- This almost satisfies the PoC milestone. However, we need completed jobs for the actual Tier-0 workflow to declare this fulfilled.
- We can expect test jobs defined by the Tier-0 team to be sent via prodsys to CSCS-PIZDAINT in the next few days. These will be jobs of the type "steady spillover O(10%) of physics_Main rate", used to commission the workflow on the grid - We refer to this as "steady spillover at lower rate".
- One job injected, waiting in the queue: https://bigpanda.cern.ch/job?pandaid=3925219511
- Rod could trigger the current test Tier-0 task "steady spillover O(10%) of physics_Main rate" to be sent to CSCS-PIZDAINT. Gianfranco will ask for this [ACTION].
- Still unclear how to scale up to 150 nodes, will try to judge from the size of the above task, but very likely we'll need a full "physics_Main" task
- It is accepted that production start will be delayed accordingly, with date to be defined.
- SSD layer is accessible (and accessed) since Monday 7th
- dCache storage (1 PB & token) are still on the works
- ARC servers (4) are still on the works.
It was agreed that all 4 servers should be used for data staging and 2 of the 4 for job submission. Publishing to the BDII and APEL should not be done in order to avoid Ops jobs from getting in and messing up with the Tier-2 accounting.
- Gianfranco:
The current architecture foresees 1 ARC CE and 3 Data Stagers. We should aim at 2 ARC CEs and 2 Data Stagers to achieve HA, as the ARC services can be a crucial point of failure. The ARC CE natively performs data staging, so we will still have 4 stagers anyway. The CE setup and hardware requirements are very similar between CE and DS. Some attention to be paid to the infosys and accounting services.
Gianfranco:
Agreed with Gianni to add the needed GOCDB entries in order to publish properly to EGI as well. Not sure how "politically critical" this is, but if this does not work, directories holding unpublished job records will fill up quickly on the ARC CE and alternative mechanism should be put in place to clean up (not ideal). Will follow up in case further entries will be needed, e.g. declare the ARC CE services in the GOCDB and/or in the site-bdii.
CSCS: deduced we don't want to report to EGI/APEL