Geomancy: Automated Data Placement Optimization

Student: Oceane Bel (University of California, Santa Cruz)
Supervisor: Darrell Long (University of California, Santa Cruz)

Abstract: Exascale cloud storage and High-Performance Computing Systems (HPC) deliver unprecedented storage capacity and levels of computing power, though the full potential of these systems remain untapped because of inefficient data placement. Changes in data access patterns can cause a system's performance to suffer. To mitigate performance losses, system designers implement strategies to preemptively place popular data on higher performance nodes. However, these strategies fail to address a diverse userbase whose users individually demand the highest performance, and they must be carefully constructed by an expert of the system.

We propose Geomancy, a tool that reorganizes data to increase I/O throughput. In systems where heuristic-based improvements may become resource intensive, Geomancy determines new placement policies by training a deep neural network with past workload and system traces. With real workload traces, Geomancy calculated an example placement policy that demonstrated a 49% increase in average throughput compared to the default data layout.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: pdf


Back to Poster Archive Listing