Public storage for the Open Science Grid

20th International Conference on Computing in High Energy and Nuclear Physics

物理学;计算机科学

Levshina, T.^1 ; Guru, A.^2

Scientific Computing Division, Fermilab, Batavia

IL

60510, United States^1

Holland Computer Center, University Nebraska-Lincoln, Lincoln

NE

68588, United States^2

关键词: Data management workflow;    Front-end services;    Local institutions;    Open science grid;    Performance tests;    Production manager;    Storage elements;    Virtual organization;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/513/3/032057/pdf
DOI  :  10.1088/1742-6596/513/3/032057

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

The Open Science Grid infrastructure doesn't provide efficient means to manage public storage offered by participating sites. A Virtual Organization that relies on opportunistic storage has difficulties finding appropriate storage, verifying its availability, and monitoring its utilization. The involvement of the production manager, site administrators and VO support personnel is required to allocate or rescind storage space. One of the main requirements for Public Storage implementation is that it should use SRM or GridFTP protocols to access the Storage Elements provided by the OSG Sites and not put any additional burden on sites. By policy, no new services related to Public Storage can be installed and run on OSG sites. Opportunistic users also have difficulties in accessing the OSG Storage Elements during the execution of jobs. A typical users' data management workflow includes pre-staging common data on sites before a job's execution, then storing for a subsequent download to a local institution the output data produced by a job on a worker node. When the amount of data is significant, the only means to temporarily store the data is to upload it to one of the Storage Elements. In order to do that, a user's job should be aware of the storage location, availability, and free space. After a successful data upload, users must somehow keep track of the data's location for future access. In this presentation we propose solutions for storage management and data handling issues in the OSG. We are investigating the feasibility of using the integrated Rule-Oriented Data System developed at RENCI as a front-end service to the OSG SEs. The current architecture, state of deployment and performance test results will be discussed. We will also provide examples of current usage of the system by beta-users.