Personal tools
You are here: Home Virgo Data Analysis VDAS vdasc_20141201
Log in

Forgot your username or password?
Document Actions


by Gergely Debreczeni last modified 2014-12-01 16:09


    1.) News, announcement
           o report on LVCComputing activities) (Gergely)
           o news from JECC meeting (Gergely)
    2.) Status and future plans of the Polinomial search pipeline (Reinier) pOLINOMIAL SEARCH
          o Reinier will present the pipeline and some profiling result and we will discuss possible alternatives on ho to 'GPUize' the hot-loops of the code.
    3.) AOB


Reinier, Luca, Michael, Andrzej, Gergely


Gergely communicated the news from JECC meeting and from LVCComputing meetings, Bari resources, and ongoing activities.

  1. JECC: New core allocation requests for Virgo for 2015 will be the following: 300 core in Lyon and 600 core additional in CNAF, so there will be in total 1000 core available at CNAF. EGO Council still has to approve this request. The unusually high request for Lyon has been put in place in order to balance better the usage of our two main computer centers. We have to do everything to make those resourece more conveniently usable for the collaboration members There will be 10 GPU in Lyon, 20 in CNAF for production from the second half of 2015 and 4 GPU in Cascina for development from beginning of next year.
  2. LVCComputing: Comittee members have been charged to contact analysis groups and collect their preference about where and when they would like to run their pipelines in 2015. After the requests have been collected the LVCComputing Comittee will check the conflicts and try to come up with a mathcmaking solution which is in a sense optimal. This will require some additional iteration.
  3. Usage of Bari Grid cluster resources is still under discussion, there is several possibility the picture will be clear later.
  4. Activities:
    1.   Lisa is working on the porting of cWB pipeline to Pegasus and execution under Grid environment. Despite the work is done she experiences some problem in connection with stability of Pegasus. Some DAG post processing capability and data transfer problem arises. Further investigation and help will be performed asking on pegasus-users list.
    2. Luca is improving Pegasus site.xml configuration in order to include all the small Grid site into the Pegasus scope, so uniform submission will be possible. Luca also works on LDR installation at CNAF.
    3. Mab is working on LDR installation at Lyon. After installation of LDRs in CNAF and Lyon a throrougful testin activity  will be required.
    4. Gergely GPU porting with Reiniert, Pegasus - CondorC job submission testing

Luca also reported that CNAF colleague XXXX is ready to help us portig code to GPUs. (His name will be advertized here.) There is also possibility to have interactive jobs in CNAF which will help a lot testing and debugging, Luca will arrange this. It will be very useful for example to Lisa to test Pegasus log files.

Reinier gave a "computing focused" presentation on Polinomial search pipeline. The algorithm and some profiling result have benn presented. The hot loops of the code (70%) was identified as the Fourier transformation which is quite easy to put to GPU. Some latency in data access also experienced (20%) and the rest is template generation (10%). Many similarities with CBC template matching code so maybe some already implemented algorithm can be taken / reused from there... will see. Some discussion on how to template generation should be done directly in Frequency domain also took place, but no immediate solution was found. Will look later. Gergely and Reinier agreed that they will have a look to the code together and discusson what the next step should be.

Similar activity and code report is to be expected from other groups in the following months.