Batched Pattern-Aware Cache Management Strategy for Astronomical Time Series Sub-images Retrieval

Astronomical research has entered an era of big data as the astronomical science data are so rich and keep increasing dramatically. A hotspot event (e.g., discovering the supernova explosion) can attract massive access to image data covering a small sky area in a short period of time, these data are just a drop in the ocean of archived data. For further study in the event, efficient acquisition of required target images is a prerequisite for specific research. Astronomical images are usually stored in the FITS (Flexible Image Transport System) format, with image sizes ranging from Megabytes (MB) to Gigabytes (GB). It is time-consuming to cut the time series subimages from the FITS images in the archived layer directly. If FITS sub-images are cached for the next query, a large amount of data access with subtle differences in target sky area range will cause frequent data caching and replacement. Therefore, we designed Batched Pattern-Aware (BPA) cache management strategy, which is optimized for massively concurrent access to the time series sub-images covering small interested sky area. Based on a hierarchical storage strategy to manage archived images and cached sub-images, the Batched Pattern Aware mechanism optimizes massively concurrent requests in the query queue within a short time. By preprocessing these concurrent requests, the correlation between the request history and the current requests are considered to reduce data erasure at the cache layer, improve the cache hit rate and shorten average response time. This strategy makes up for the shortcoming of considering only long-term historical requests and ignoring the current request status. Experimental results show that the proposed BPA strategy can achieve a higher cache hit ratio among 9 algorithms, the average response time can be reduced by at least 46.30% comparing with the others.