Skip to main content
. 2021 Nov 29;18(12):1496–1498. doi: 10.1038/s41592-021-01326-w

Fig. 1. Chunk retrieval time is less sensitive to data location with next-generation file formats.

Fig. 1

a,b, Random sampling of 100 chunks from synthetically generated, five-dimensional images measures access times for three different formats on the same file system (green), over HTTP using the nginx web server (orange) and using Amazon’s proprietary S3 object storage protocol (blue) under two scenarios: a whole-slide CycIF imaging dataset with many large planes of data (x = 64,000, y = 64,000, c = 8) and chunks of 256 × 256 pixels (128 KB) (a); and a time-lapse LSM dataset with isotropic dimensions (x = 1,024, y = 1,024, z = 1,024, t = 100) and chunks of 32 × 32 × 32 pixels (64 KB) (b) (Methods).