1. www.bsc.es LOFAR - Calibration, Analysis and Modelling of Radio-Astronomy Data EGI Conference 2015 19 May 2015, Lisbon Daniele Lezzi – Barcelona Supercomputing Center Susana Sanchez-Expósito - Instituto de Astrofísica de Andalucía – CSIC José Sabater - University of Edinburgh

2. The LOw Frequency Array (LOFAR) Credits: ©ASTRON. https://www.astron.nl/ Radio interferometer composed by thousands of antennas distributed among several countries Application: Astronomy Galaxy CIG292. Image from SDSS Calibration pipeline use case: Pipeline for transforming the raw data from the antennas to science-ready data Existing collaboration in the AMIGA4GAS project This pipeline would be use by tens of users Calibration pipeline use case: Pipeline for transforming the raw data from the antennas to science-ready data Existing collaboration in the AMIGA4GAS project This pipeline would be use by tens of users

3. Why to use Fedcloud Frequency Sub-band 1 Sub-band 2 Sub-band n X GB DATA PARALLELI SM Needs to process large datasets (several TBs) Datasets can be divided in pieces that can be processed independently. COMPSs useful for automatic scaling and elasticity features. Up to 488 independent tasks working on dataset pieces of several GBs Further details in poster titled: EGI Federated Cloud for calibrating and analysing Radio-Astronomy data

4. Development use case status Pipeline modelled in COMPSs, that provides a Python binding and takes care of the parallel execution of the tasks. Virtual Appliance customized with the LOFAR software and deployed through the AppDB. import subprocess import sys import os from pycompss.api.task import task from pycompss.api.parameter import * @task( script_name = FILE) def iter_calib(script_name): os.chmod(script_name,0744) subprocess.call(script_name) print "end executiong" if __name__ == "__main__": args = sys.argv[1:] DATA_PATH=args[0] TEMPLATE_FILE=args[1] f=open(TEMPLATE_FILE,'r') content=f.read() f.close() list_f=os.listdir(DATA_PATH) for directory in list_f: # Iterate over the data inputs if os.path.isdir(DATA_PATH+"/"+directory): new_content=content.replace("INPUTDATAPATH",directory) script_name="job"+directory+".sh" f=open(script_name,"w") f.write(new_content) f.close() iter_calib(script_name) Integration with the Fed Cloud through COMPSs for VM management (using rOCCI). The data is uploaded on a NFS storage (on each site) and shared amongst VMs. No object storage used.

5. User Experience/Wish list The AppDB Virtual Appliances facilitate the installation of the user application on Fedcloud. The users can directly install their software on a running Fedcloud VM, which has been previously started as an instance of a Virtual Appliance. The process of creation and deployment of the Virtual Appliances in the AppDB is not automatic. There should be a way to let users create snapshots. New software releases imply starting again the deployment process. Last year LOFAR software team published 8 releases. The computing power provided by the different sites federated in Fedcloud allows to configure virtual machines with different capabilities (memory+cpu) in order to match with the specific needs of the use case. Lack of consolidated solutions for federating storage. The user data are too large to be stored in the VM images. They should be stored in volumes easily mountable from several VMs and synchronized across different sites. The COMPSs framework facilitates the porting and deployment (through the PMES service) of the application. COMPSs allows the users to program their applications sequentially, in a transparent way by taking care of their parallelization as well as of the VMs orchestration for their execution. User software deployment Scalability / performance User interaction Friendly tools for working as interface for the command line client would facilitate the first steps in using Fedcloud: to configure the VMs and install the software, manage the proxy certificates and propagate the new releases of Virtual Appliances to the Cloud Marketplace. Blocker!!

6. Future plans Continue and refine the implementation of the use case. More scalability and elasticity tests with multi-providers runs. Integration with a better data management solution. Evaluate PMES as end users interface. Integrate the application in the Astrophysics Science Gateway (see demo #2)

7. www.bsc.es Thank you! http://amiga.iaa.es/p/263-federated-computing.htm http://compss.bsc.es