When I/O Workloads Don’t Perform

Other Things to Consider

You will want to know whether you are looking at a solid state drive (SSD). If so, performance can drop drastically once the SSD reaches its write cliff, or the point at which the drive exhausts its first writes to each NAND memory cell. Once you reach the drive’s write cliff, you are now exercising a programmable erase (PE) cycle for each write operation. This multistep process involves erasing an entire cell before writing to it. Each NAND cell has a finite number of PE cycles. The PE cycles to a cell will vary on the type of NAND technology employed. Vendors resort to something called over-provisioning to prolong or delay this write cliff.

Often you will get double the NAND memory when purchasing a drive. For instance, on a 3TB PCIe SSD, you might be paying for 6TB of storage, of which you are only able to use 3TB. There are two methods by which someone can restore an SSD to its former performing glory. One is by employing a technique called TRIM (called Discard on SAS technologies). In this background process the software tells the drive to prep unused data blocks for a future write operation. Tools such as fstrim will enable this function, but many filesystems (e.g., ext4) and volume managers (e.g., logical volume management, LVM) offer discard support. Although a bit more data destructive, it is a hard format operation at the drive level.

Conclusion

When something does not seem quite right, you have many ways to peek into a system – a few of these mechanisms were explored here. This guide does not cover everything and just provides a starting point, but it is enough of a starting point to get you moving in the right direction of isolating and diagnosing the root cause of your performance troubles.