Large-Scale, Multi-Sensor Atmospheric Data Fusion Using Hybrid Cloud Computing

NASA's Earth Observing System (EOS) is an ambitious facility for studying global climate change. The mandate now is to combine measurements from the instruments on the "A-Train" platforms (AIRS, MODIS, MLS, and CloudSat) and other Earth probes to enable large-scale studies of climate change over decades. Moving to multi-sensor, long-duration presents serious challenges for large-scale data mining and fusion. For example, one might want to compare temperature and water vapor retrievals from one instrument (AIRS) to another (MODIS), and to a model (ECMWF), stratify the comparisons using a classification of the "cloud scenes" from CloudSat, and repeat the entire analysis over 10 years of data. HySDS is a Hybrid-Cloud Science Data System that has been developed and applied under NASA AIST, MEaSUREs, and ACCESS grants. HySDS uses the SciFlow workflow engine to partition analysis workflows into parallel tasks (e.g. segmenting by time or space) that are pushed into a durable job queue. The tasks are "pulled" from the queue by worker Virtual Machines (VM's) and executed in an on-premise Cloud (Eucalyptus or OpenStack) or at Amazon in the public Cloud or govCloud. In this way, years of data (millions of files) can be processed in a massively parallel way. Input variables (arrays) are pulled on-demand into the Cloud using OPeNDAP URLs or other subsetting services, thereby minimizing the size of the transferred data. We are using HySDS to automate the production of multiple versions of a ten-year A-Train water vapor climatology under a MEASURES grant. We will present the architecture of HySDS, describe the achieved "clock time" speedups in fusing datasets on our own nodes and in the Amazon Cloud, and discuss the Cloud cost tradeoffs for storage, compute, and data transfer. Our system demonstrates how one can pull A-Train variables (Levels 2 & 3) on-demand into the Amazon Cloud, and cache only those variables that are heavily used, so that any number of compute jobs can be executed "near" the multi-sensor data. Decade-long, multi-sensor studies can be performed without pre-staging data, with the researcher paying only his own Cloud compute bill.