Why would you randomly access the data? Instead of reading the data randomly or in some kind of random-access inducing order, you'd be much better off scanning through the data sequentially by sector or extent to get data in as large of contiguous chunks as possible. Granted with SSD and NVMe drives it doesn't really matter. My ~29K message work inbox shows an average message size of 63KB. At that size solid state storage is already throughput bound rather than IOPS bound. A modern NVMe drive like a P3700 should be able to read that data in it's entirety in about 20 seconds assuming the CPU can keep up. If the FBI is using a reasonably decent distributed storage system that distributes objects across multiple device you can easily do it far faster. ~20GB/s and half a million read IOPS is nothing special for distributed storage systems on NVMe.