The FRIB data acquisition working-group met at the Physics Division of the Argonne National Laboratory to discuss data acquisition needs with representatives of the FRIB community. The representatives were appointed by the FRIB working groups to present a summary of their intentions and needs for data acquisition support.
The first day of the meeting consisted of presentations by the working group representatives. Members of the NSCL staff made presentations on the computing and data acquisition infrastructure and possible future evolution. Finally, experimenters from several institutions presented their experiences in integrating multiple, complex detector systems into a single experiment, including discussions of event building and time synchronization.
The second morning of the working group meeting was devoted to discussions of the areas in which the community perceived needs for development in data acquisition. The discussions were framed around:
Then the discussion was directed toward identifying new developments needed to support questions raised during the discussion, and the involvement of the community in those discussions.
We first discussed developing community protocols and interfaces to support experimenters develop data acquisition applications. These commonly shared and supported practices will help experimenters more easily interface their systems, large and small, to other detectors, accelerator resources, facility computing and storage resources, and provide reliable control and event identification. We can even use them to develop facility resources to implement the protocols and interfaces. The areas for development we identified in this meeting are:
Some members of the community have already volunteered to assist in these tasks.
Our second discussion focused on data analysis. Participants expressed concerns about having sufficient resources to analyze experimental data during the measurement, and to perform offline analyses of the merged dataset in the same or less time than was required to make the measurement. In addition, there were concerns about the short and long-term storage of data including access to it on FRIB-provided storage.
We live in a time of rapidly expanding capabilities for data processing, often labeled as "Big Data", yet few, to our knowledge, in the nuclear physics community are taking advantage of these technologies and techniques. Experimental systems like GRETA and the AT-TPC can generate huge amounts of data, for which analysis can take a long time with the usual resources available in the community. The participants decided that there should be a "Workshop on New Methods in Data Analysis" to explore how new computing resources and techniques associated with Big Data can be applied to our data analysis, enabling analyses that look deeper into the data and in shorter processing time. These modern data analysis methods need to be reliable, provide high throughput and low latency, be easily usable and standardized enough that new applications are straightforward.
The issue of resources for data storage and analysis came up often. There was a need expressed for experimental event data to be retained for a long enough time on servers at FRIB. FRIB datasets will be multi-terabyte in size, driven by flash ADC acquisition and the growing complexity of experiments. Moving these data over the Internet to researchers' home institutions or to external data processing resources will demand very high bandwidth connections to the Internet, for example to access resources such as Globus, which provides very high bandwidth, asynchronous transfer of data to and from large data centers. If the network bandwidth to the experimenters' home institution is limited, processing resources at FRIB will be required to help in data analysis. This is also very important for some experiments with high latency event building, like the Decay Station, or complex reconstruction, such as the AT-TPC, where significant resources are needed to understand the functioning of the experiment online. High priority, high throughput computing is needed for experiments as these, available online. We must investigate more to identify in detail the needs of the community for FRIB local processing.
To address these topics, we will form a data analysis special interest group.
The FRIB data acquisition working group will oversee the operation of the panels. They will:
We plan to have an FRIB DAQ working group member act as the chair and point of contact for each panel, with regular (quarterly or monthly) reports to the WG as a whole. To keep the efforts focused and to not overburden the community, we will startup only one or two subgroups at a time. Some will resolve their discussions quickly, others may take a few years of development.
We are preparing a report of workshop for FRIB management, to begin discussions about these issues in their planning for FRIB operations.
We will organize an annual meeting of the FRIB data acquisition working group, similar to that we just had, to discuss progress on current issues being investigated, to publicize completed work and learn about new issues.
We plan to have the first analysis workshop in about a year. We encourage the community to investigate Big Data techniques and be prepared to come describe experiences with everyone.
The first two panels, Time and Accelerator/Beamline interface and controls will be organized and in operation by September 11, 2015.
We welcome those who are interested in FRIB data acquisition to join the working group or join a topical panel. The community's input to this is critical. Consensus can only happen from community effort.
Please send an email to Robert Varner (varnerrl@ornl.gov) to join, or to participate in one of the subgroup activities.