Automatic via web iface
- Go to http://portal-auth.nersc.gov/alice and find a link
New request. - Once you klick there you open a new form where you enter
periodreconstruction passdata type(ESDin case of ESDs,AODorAODxfor AOD productions) and arun numberor alist
Behind the scenes
The script for manual download will be handled by a cron job on dtn1 node. With this it is also possible to transfer several runs in one shot - tested with two in parallel.
Note that the cron jobs processing the request, distributing the jobs and launching them have timeouts of 5 min each - it can take up to 15 min until a job gets started.
Prepare transfer using scripts
NOTE: Needs user to be in aliprod group
How this works...
- The whole stuff is under /projecta/projectdirs/aliprod/data/work
- The system automatically distributes load to the different dtns
- The token is renewed automatically. It goes under my name, using a long-lived proxy (the current one will expire in November)
- There is a small monitoring: http://portal-auth.nersc.gov/alice . For more we are lacking the mount of projecta on portal-auth. Due to the last interventions it is currently down - please give me 5 minutes to launch it again.
Request a transfer
In order to create a new transfer job you need the script create_copy_filelist.sh, It takes the following parameters:
- inputpath on ALIEN (
/alice/data/2016/LHC16n) - outputpath on /projecta (
/global/projecta/projectdirs/aliprod/data/LHC16n); keep in mind that the dtns do not have the symlink /projecta to /global/projecta - run number (
260710) - pass (
muon_calo_pass1) - optional (Filetype and archive - needed for AOD transfer, automatically handled for ESDs)
The script will produce a textfile inside your pwd. This textfile needs to go to the directory queued. A cronjob on dtn1 will check for new jobs every 5 minutes and assign to a worker (as long as there is at least one worker with less than 5 transfers ongoing). Internally the transfer client is threaded (default: 10 threads), meaning that we are transferring 10 files in parallel.