Seek times? multiple simultaneous requests for data access? why was NCQ even implemented if these were not a problem for spinning media?
Seek time (or latency as it's often called) is its own thing, even though it's closely related to IOPS and throughput. Generally speaking, an increase in latency means a drop in IOPS (the added latency increases the completion time of the IO request, which in turn decreases the IOs per second). However, the formula for IOPS is still:
IOPS = [Bytes per second] / [IO size in bytes]
Throughput is Bytes per second, so the only difference is that IOPS does not take the IO size into account, although IOPS is always reported for a specific IO size (usually 4096 bytes as manufacturers tend to give out 4KB random read/write figures).
Just try accessing 3 or 4 (or more) files at the same time with a hard drive, then do so with an SSD, and you'll see the difference between IOPS and throughput.
The biggest difference there is seek times. When you're accessing multiple files simultaneously, the drive has to access LBAs in a random order (e.g. one file is using LBA range from 1 to 20 and the other file is using 400-420). Hard drives suck at this because the disk head has to constantly jump from one place to another and you're limited by the drive's rotation speed. Since the latency to access one LBA is long, the IOPS is also poor (hard drives can barely achieve a few hundred IOPS in 4KB random read/write).
SSDs have very low latency (read/write latencies are measured in microseconds) and don't care if the IO requests are coming in random or sequential order because they are not limited by mechanical movement. Hence it's not an issue to access multiple LBAs simultaneously as the latency to access one of them is so short and with individual IO latencies being very short, an SSD can complete tens of thousands of IOs per second.
Facebook
Twitter