Unit for Processing and Data Archiving
The data archiving and processing of the images collected at the OAJ will be carried out in the Unit for Processing and Data Archiving (UPAD). This data center will provide the hardware infrastructure needed to store, process and analyze the images, as well as keep data backup. It will also have the provide efficient access to the scientific database and sky images for the astronomical community and the general public.
The UPAD hardware has three main systems:
- UPAD/T250 main storage and processing
- EDAM (External Data Access Machine)
The system has two dedicated IO servers that control the movement of the data from the camera servers to their mirrors (other two IO servers) at the UPAD. During day time those two servers secure two backup copies of raw data. The OAJ has a storage system 90 TB net that acts as a buffer for two months of data collection. The project status can be found here.
UPAD storage and processing
At UPAD datacenter, the main logistic about data handling and pipeline processes scheduling is controlled, apart from the two IO servers dedicated to the download of OAJ data and service nodes, by two queue control servers that run Sun Grid Engine as batch queuing system. An scheme showing the nodes and storage systems deployed at OAJ and UPAD is shown in figure 1.
UPAD main storage and processing capabilities
The hardware for storaging and processing the J-PLUS data collected with JAST80 is already deployed. It consist on 2 processing nodes, apart from the queue control servers, 2 database servers and a centralized disk storage system with a capacity of 90 TB net.
Concerning the data produced by the JST250, the UPAD storage system has two main functions. On the one hand, it has to store a enormous volume of information, only the J-PAS raw data amount to ~ 1 PB after compression, and each complete data release is ~ 850 TB. On the other hand, the centralized storage system shall provide the IO rate required for hundreds of processes executing the pipeline processes for different input frames. In order to fulfill those two requirements and allow the increasing storage demands, the UPAD combines disk and tape technologies. The disk storage system shall have the capacity to store the hot data that is accessed by the pipelines, providing a high bandwidth in order to feed all the concurrent pipeline processes. The robotic tape library acts at a near-line storage for all the products with lower access frequency, for example all the raw data and earlier Data Releases.
The main UPAD storage system deployment was carried out during December 2014. The disk storage system consists on a Netapp cluster with 8 nodes providing a net storage capacity > 1000 TB with dual parity protection. The robotic tape library Spectra Logic T950 with 2 frames has 1600 LTO6 slots (~4 PB). Both storage tiers are integrated by a HSM solution. The core network and disk storage system provide more than 5000 MB aggregated bandwidth. The global storage systems and core network solution has been designed and will be integrated and deployed by BULL ESPAÑA.
Concerning the processing systems. In order to minimize the IO operations overhead, the pipelines can be configured to store all the intermediate product in RAM drive. Also the IO to the centralized storage is decreassed by storing locally, in the computed nodes, some frequently accessed data as for example Calibration Frames.
An important part of the computing resources at UPAD was deployed during the year 2015. This first processing infrastructure consists on 17 Fujitsu servers. Each server has 2 CPUs with 12 cores, 192 GB of RAM, and 4.0 TB of scratch storage.
The UPAD storage, network, and processing infrastructure is funded by the Subprograma de Proyectos de Infraestructura Científico-Tecnológica of the Spanish Ministry of Economy and Competitiveness (MINECO) (FCDD10-4E-867), cofunded by the European Fund for Regional Development (FEDER) and Fondo de Inversiones de Teruel (FITE).
Software pipelines have been designed to handle the enormous data flow produced by the panoramic camera and maximize the scientific output. The Data Management Software will automatically process the data collected during the night to check if its quality fulfills the scientific and technical requirements, update the survey's databases and feed the Scheduler to compute the telescope targets of the following nights.