Hardware Recommendations for Faster Processing
Published On: November 08, 2016
If you're i/o bound, which is usually the case with LIDAR or photogrammetric point cloud processing, you need to figure out what kind of i/o you're doing at the bottleneck.
Case 1: If you're just waiting for a single tile to load, your slowdown may be from low RPM HDD, but more likely you're hitting the SATA bus limits. You won't see much improvement by switching to SSD unless it's a PCI express m2 SSD, which has a much faster data bus.
Case 2: If you're sampling pieces of multiple files (e.g., processing buffered tiles with neighbors, processing sections of tiles), then you'll see a big improvement processing from SSD because of faster seek times. Consider a SSD RAID array. Caution: Be sure to select the right type of disk because of how the RAID card read/writes to it.
Case 3: If you're running multithreaded/multiprocess and multiple threads are reading/writing at the same time to the same drive (e.g., multi-process spawning, uses multiple cores), you might be running into thrashing. This can be reduced by running on SSD arrays, but the better option is to modify your process to ensure only one thread/process is reading or writing to a specific drive. For example, if running multiple instances of a software, dedicate a drive to each instance.