Products related to this article
＊This article is a revised version of the article posted to Qiita in November 2019.
Introduction
In the previous article, we explained that TBW indicates only the data newly written from the host, while the estimated lifetime indicates all the data written to the NAND flash memory in the SSD.
If factors other than TBW that consume the life of the NAND flash memory becomes larger, as a result, TBW numbers will get smaller.
The size of the above factors depends greatly on the characteristics of the product, so it is not always possible to say that “large = bad”. However, it is possible to prevent excessive increase in the size of the product by devising how to use it.
Thus, from this article, we will explain the bad way that shortens the life of SSDs (i.e., diminishing TBW) .
By avoiding the usage described in this article, you can maximize TBW and take full advantage of the life of your SSD.
Summary
- Do not use without enough capacity
- Random writes to a large LBA space should be avoided
- Continuously reading the same data is also a problem
No room for capacity is not good!
＜Point: If it is used with sufficient capacity, it enables to keep stable performance of SSD.＞
NAND flash memory cannot write data unless it is erased block by block. Therefore, SSDs reuse blocks to create writeable blocks before those blocks run out, thereby maintaining storage functionality.
Garbage collection (GC) is the process of sorting out valid and invalid data in a block and copying only the valid data to another block for the purpose of reusing.
Under certain assumptions, when comparing the case where the effective data volume is 50% of the SSD capacity with the case where the effective data volume is 80% of the SSD capacity, the latter consumes 3.2 times more lifetime during GC! (See Qiita article for details).
For this reason, SSDs should be used with ample capacity. And the easiest way to make room for capacity is to select larger capacity.
For example, even if the size of the data expected to be stored about 100 GB, choosing a 256 GB SSD instead of a 128 GB SSD is a much gentler usage.
No random writes to a large LBA space!
＜Points: Much higher performance than HDD, but not good random writes＞
Compared to the days when the performance degradation known as “petit freeze” was observed, SSD random write performance, especially for small size (e.g., 4 KB) random writes, is now very high.
One of the reasons for this is the difference in the method used to manage the correspondence between the address (Logical Block Address: LBA) specified by the host for reading and writing data and the location of the NAND flash memory where the latest data (LBA) is recorded.
The management method used for most SSDs in recent years, including our SSDs, is called “page mapping” or “page-level mapping[1][2][3].
This page-mapping scheme enables small-sized writes to be efficiently recorded and managed in NAND flash memory, especially improving random write performance. In addition, this method is compatible with the OS, especially the Windows access pattern[4], and performance in an actual environment has also been greatly improved.
The previous management method is called “block mapping” or “block-level mapping” because it manages NAND flash memory in block (size) units. There is also a method called “hybrid mapping” that combines these two management methods.
However, even with the page mapping method, the GC becomes inefficient if random writes are performed over a large area of LBA space. Also, under such circumstances, the bloated management information which is a disadvantage of the page mapping method, will affect the lifetime of the system.
Efficiency deterioration in GC means an increase in copy size of the GC that is to say an increase in lifetime consumption, resulting in a larger decrease in TBW. In other words, random writes (especially small writes) over a large area of LBA space are not SSD-friendly.
On the other hand, sequential writes require little or no GC or can be performed relatively efficiently, resulting in less TBW depletion.
Don’t let your devices keep reading the same data!
＜Point: Be careful with workloads that frequently read specific data.＞
NAND flash memory has a function called Read Disturb.
The increase in the number of errors due to this characteristic should be addressed. For example, by recording the number of readings and copying the data to another location when the threshold is exceeded. Our SSDs are equipped with refresh as a countermeasure against this read disturbance.
However, this process also causes data copy(writing to NAND flash memory) which increases lifetime consumption and reduces TBW.
In other words, the use of SSDs to read the same data a lot of times in a short period is not NAND-friendly.
The frequency and number of reads for which this characteristic becomes a problem depends on the NAND flash memory and the management algorithm of the SSD.
In our SSDs, the characteristics of the NAND flash memory used are confirmed through in-house experiments, and the number of read cycles to perform the above refresh and other parameters are set.
Conclusion
Lastly, this article summarizes the “not-so-gentle” use of SSDs from the perspective of its lifetime consumption.
There are a wide range of using SSD which can affect its lifespan, however, we hope this will give a hand for you when selecting an SSD in the future.
The next article in this series, we will continue explaining the “not-so-gentle” use of SSDs.
References
[1] Kim, et al., “A Space-efficient Flash Translate Layer for Compactflash Systems”, IEEE Transactions on Consumer Electronics, vol. 48, no. 2, pp. 336-375, May, 2002
[2] Hu, el al., “Achieving page-mapping FTL performance at block-mapping FTL cost by hiding address translation”, in proceedings of IEEE 26th Symposium on Mass Storage Systems and Technologies (MSST), May, 2010
[3] “A Summary on SSD & FTL“ ,Viewed June 14, 2022
[4] Aso, Hanabusa, “SSD Mechanisms and Characteristics,”Design Wave Magazine, no. 132, page 36, October, 2008
Trademarks of Other Companies
Although registered trademark marks are not indicated in the articles, company names and product names appearing in the articles are generally trademarks or registered trademarks of the respective companies.
About the article
The content of this article is information at the time of publication. Please note that the information is subject to change without notice.
