In the storage industry, storage performance testing is naturally indispensable. There are generally two methods for performance testing:

1 Use industry recognized performance testing tools such as IOMeter, fio, iozone, etc. These types of tools are generally open source and have a wide range of applications. One advantage of using them is that they can quickly compare horizontally with other storage products. Even if everyone tests at different times, locations, and environments, as long as the parameter settings are similar, a simple comparison can be made (if a complete comparison is to be made, it must be tested in the same environment). The disadvantage is that they can only be mechanically tested according to predefined patterns, and in most cases, they cannot truly reflect the IO patterns of the actual application environment 2. Conduct performance testing directly in real application environments. The results obtained by this method are authoritative and reference specific to specific scenarios. The downside is that setting up an environment may be cumbersome, as each application's dependent environment is different. Even the same storage device has significant performance differences in different application scenarios. So this method can reflect the actual performance of storage products in specific scenarios, but sometimes it cannot determine the general performance of storage products When using the same storage device and hard drive, the performance difference may be significant when the following conditions change 1. Read and write. In theory, the sequential read and write performance of mechanical hard drives is similar. But in different scenarios, the read and write performance of the same hard drive varies. Because caching plays a greater role when writing data, writing performance may be better. But if RAID5 is used, because there is a write penalty for writing data, the performance of reading is better when other conditions are the same. The read and write parameters of SSDs are inherently different, as writing data can cause block erasure, resulting in faster reads, especially random reads 2. Sequential IO and random IO. For mechanical hard drives, there is no doubt that sequential read and write is much faster than random read and write, as sequential read and write reduces a lot of seek time and its speed is 1 to 2 orders of magnitude higher than random read and write. Therefore, traditional storage requires the use of caching algorithms (including sorting by sector address and pre reading) to reduce seek time and improve performance. For SSDs, the difference between sequential and random reads and writes is not that significant, but sequential reads and writes are generally faster than random reads and writes. In practical applications, pure random reads and writes are rare. Even when copying large files in a file system, there are still some relatively random operations, such as modifying metadata and skipping occupied sectors. So, how to simulate the randomness of an application is a headache