IBM Power System 9040-MR9 Manual page 55

written (programmed) and read in small units that are called pages, individually, but to rewrite that page,
the page must first be erased and then programmed again.
Due to the architecture of NAND flash, erase operations are completed on the block level and not the
page level. Each block contains hundreds to thousands of pages. Therefore, to erase a block, all of the
valid data must first be transferred to another block and then you can erase the block. The SSD controller
then searches for blocks with large proportions of pages that have data that can be erased. Then, the SSD
controller moves and combines the pages of data that must be retained to previously erased blocks,
which then frees these new blocks for erasing.
This process of moving data to free up blocks for erasing is called garbage collection. Increasing the over-
provisioning of an SSD enables the controller to be more efficient in garbage collection and minimizes
extra read and program operations.
All of these background operations result in more data that is written to the flash than what is written to
the drive. The ratio of data that is written to flash and the data that is written to the drive is referred to as
the write amplification. All else being equal, write amplification is greater for drives with lower over-
provisioning.
Lower cost for mainstream drives
The cost per GB of a mainstream drive is typically less than the cost per GB of an enterprise drive. The
cost is low because the amount of over-provisioning is less and almost all the flash memory in the drive is
available for storing data.
Lower write operation performance for mainstream drives
Mainstream drives are similar to enterprise drives in terms of read performance. However, because of the
lower over-provisioning of the mainstream drives, the random write performance is reduced because of
the higher number of background operations that are needed for garbage collection and associated write
amplification. Thus, lower over-provisioning decreases both performance and endurance. Read
performance is not impacted.
No mixing of mainstream drives with enterprise drives in disk arrays
Endurance differences imply that when you form disk arrays, you must not mix mainstream drives with
enterprise drives because the PCIe SAS adapter stripes data across drives, thus sending equal data to
each drive. The IBM PCIe SAS adapters do not allow mixing of mainstream drives with enterprise drives
when you create RAID arrays.
Monitoring end of life for mainstream drives
You must monitor end of life symptoms for mainstream drives because of their endurance limitations.
Internally, when the drive nears the end of life, a predictive failure analysis (PFA) trip is generated and an
operating system message is logged. When this trip is generated, the drive continues to run but it must be
replaced as soon as possible. The PFA trip code for end of life is the same as the PFA trip code for thermal
failures. Therefore, you can determine the root cause of a failure by using operating system support that
is provided by a fuel gauge command.
Warranty and maintenance information for mainstream SSDs
A mainstream drive is not suitable for write intensive workloads. Assuming a typical heavily random
workload, at about 3394 TB of write operations to a 1.9 TB mainstream drive, the drive is at its maximum
projected write capability. If the write operations exceed the drive's maximum write capacity, the write
operation takes more time to complete. A predictive failure analysis (PFA) message indicates that you
must replace the drive.
If you ignore the PFA message and if you continue to send the write operation requests to the drive, the
drive is unable to accept write commands and accepts only read commands for some time. A failed to
write operation results in a more serious error message that indicates that the drive must be replaced.
The nature of the workload impacts the maximum write operation capacity. For example, if a high
percentage of sequentially oriented write operations is used instead of random oriented write operations,
the maximum write operation capacity increases. You must periodically check the percentage of the
Disk drives or solid-state drives for the 9040-MR9 system 41