Download presentation
Presentation is loading. Please wait.
Published byMaurice Griffin Pope Modified over 6 years ago
1
YongPyong-High1 2015 Jan We appreciate that you give an opportunity to have this talk. Our Belle II computing group would like to report on Distributed computing and data handling at Belle II experiment in this talk.
2
we want to introduce Belle II experiment, computing model, distributed computing and data handling and MC production campaign.
3
Belle II detector KEKB accelerator has 3Km circulation and supply 8.0 GeV electron and 3.5 GeV positron beams for Belle II.
4
Kobayashi-maskawa won a Nobel Prize from Belle result, which is CP violation measurement, at Furthermore, Belle will be upgraded to Belle II searching the New Physics from precision measurement. Belle II will reach 50 ab^-1 in It is corresponding with 50 times of Belle.
5
Belle II collaboration is joined in 530 members of 94 institutions and they come from 23 countries.
6
Our Computing group of Belle II designed the computing model as you can see this page. The model is formed of 3 stages. First, KEK site will be produced the raw data from detector. Then the raw data will be saved in tape and storage. This data will be processed to make refined data, mdst. In addition, PNNL, USA site will back raw data up. Second, mdst will be transferred into grid site. Also, grid sites will generate Monte carlo and store it to be processed by end-users. Last, Processing results, ntuples will go to local resource, user interface for analysis. Optionally, we consider Cloud system for generating large MC production.
7
This is Belle II distributed computing system, This system is constructed with 4 dirac server, 3 AMGA server, LCG sites from Europe and OSG sites from USA as you can see this page.
8
To realize the computing system, we developed “gbasf2” for distributed computing and data
handling. Gbasf2 is the commandline client for submitting grid based basf2 jobs. It is based on existing, well-proven solutions plus extensions for Belle II. Gbasf2 is constructed with DIRAC for job management, AMGA for meta data catalog and CVMFS for software distribution.
9
Gbasf2 is also based on projects and dataset is software for job distribution. Datset is a collection of basf2 data which are stored in output files. Project is a series of gbasf2 jobs to handle data set. In a project, gbasf2 can read input dataset and generate output dataset. Gbasf2 has some functionalities as job submission, job monitoring, rescheduling and etc. you can see the commands and results in project level at right side picture.
10
We are still developing the metadata schema based on projects and dataset. Our goal is to book the data information automatically into metadata system and LFC. A dataset registeration is required for Belle II data analysis on the distributed computing.
11
Our computing group applied AMGA in gbasf2, a part of data handling, to improve the scalability and performance of metadata catalog. Left side picture shows the Data handling system and workflow of gbasf2. Right side picture shows how AMGA is worked in gbasf2.
12
In addition, Korean group developed AMGA manager that aims at providing an interactive exploration and searching environment for metadata in an user-friendly manner, and hiding complexities for accessing Grid service. It allows users to manipulate metadata schema, entries, attributes, access control, constraints, user and group information through a user-friendly GUI.
13
Finally, We have the big tests, two MC production Campaign
Finally, We have the big tests, two MC production Campaign. First MC production Campaign was taken from February 28th to March 19th this year. We performed event generation and detector simulation in 1st stage. Reconstruction was performed with 240k jobs corresponding with 60M events and 190TB of output data in 2nd stage. We found around 20% failure rate from metadata registeration, input data download and application errors.
14
Second MC production Campaign was taken in this summer
Second MC production Campaign was taken in this summer. We run Simulation and reconstruction with background mixing. In 2nd stage, we performed 630K jobs and obtained 560M events and 8.5TByes of output data. Dramatically, we reduce the failure rate around 10% to 1%.
15
We want to introduce Korean group activity, AMGA system
We want to introduce Korean group activity, AMGA system. It is operated well as a part of gbasf2 in MC production campaign as you can see the picture.
16
Also, Our Korean site, KISTI resource give a contribution for MC production campaign even if it is small.
17
Belle II still search for New physics with 50 times more data than current B factories.
We designed computing model with projects and datasets. Belle II computing group developed “gbasf2” for distributed computing and data handling based on Dirac, AMGA and CVMFS. We had two MC production campaigns this year. In two tests, Belle II distributed computing system works well. Thanks a lot to technology and resource providers!
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.