Open XDMoD (XD Metrics on Demand) and DrAFTS (Durability Agreements from Time Series) will be used for Aristotle metrics collection, analysis, and graphical display. These tools will enable the users and project management team to quickly display a wide variety of job accounting metrics over any desired time frame, as well as trend analysis results which will be used to predict when it is appropriate to burst. All three sites will run Open XDMoD instances for internal use; the portal will present a single federated Open XDMoD instance to the outside world.
Work is currently underway by UB and UCSB to add cloud metrics to Open XDMoD (see Federated XDMoD to monitor affiliated computing resources). DrAFTS will be exploited to make on-line forecasts of future performance and cost levels available to Aristotle. DrAFTS services will integrate its predictive capabilities with other metrics generated by Open XDMoD (bandwidth, VM duration, storage load, etc.) in an attempt to generate statistical bounds on guaranteed delivered performance levels.
DrAFTS will also be used to predict future load levels and the AWS spot market. Thus a user of the cloud federation will be able to determine when it is essential to burst to the public cloud (as opposed to waiting for resources to become available) and what the best price-performance product will be when bursting is warranted.
Early testing of DrAFTS on XSEDE resources by the Aristotle team demonstrated that DrAFTS is a useful tool in predicting batch HPC job wait times.
See publications for additional information.