I/O containers: Managing the data analytics and visualization pipelines of high end codes

Lack of I/O scalability is known to cause measurable slowdowns for large-scale scientific applications running on high end machines. This is prompting researchers to devise 'I/O staging' methods in which outputs are processed via online analysis and visualization methods to support desired science outcomes. Organized as online workflows and carried out in I/O pipelines, these analysis components run concurrently with science simulations, often using a smaller set of nodes on the high end machine termed 'staging areas'. This paper presents a new approach to dealing with several challenges arising for such online analytics, including: how to efficiently run multiple analytics components on staging area resources providing them with the levels of end-to-end performance they need and how to manage staging resources when analytics actions change due to user or data-dependent behavior. Our approach designs and implements middleware constructs that delineate and manage I/O pipeline resources called 'I/O Containers'. Experimental evaluations of containers with realistic scientific applications demonstrate the feasibility and utility of the approach.