Four Steps to Extend SSD Industrial Storage Service Life

SSDs provide incredible performance for storage systems, and the performance of industrial storage is 10 times higher than the fastest mechanical hard drives. However, since the write performance of the SSD will be affected by the block erase delay, the SSD is usually used as a reading device. But this is not to say that SSD does not improve the write performance. In fact, it will improve the write performance, but the write performance of the SSD is only half of the read performance. In addition, the write performance of SSD is gradually decreasing, which is also a performance problem.

SSD technology limits the life cycle of write operations (read operations will not affect the life cycle and reliability of SSD products). SSD chips use "units" to record data bits. When these cells are written, erased or rewritten, these operations will gradually reduce the performance of the cells. After a certain number of program cycles/erase cycles, a cell will be damaged and can no longer be used. The controller will record these damaged units so as to prevent these bad blocks from being used again, which is the same as what they do with traditional hard disk bad blocks. After the unit is damaged, the capacity of the SSD will decrease until it must be replaced. IT managers have invested a lot of money on SSDs, so they hope to adopt some methods to ensure the maximum service life of SSDs. SSD manufacturers will provide warranty services, and some manufacturers will even provide an additional 20% of the storage capacity to offset the impact of damaged units.

There are three main technologies for SSD manufacturing: Multilayer Cell (MLS), Enhanced Multilayer Cell (eMLC) and Single Layer Cell (SLC). MLC is a product for individual consumers, because its manufacturing cost and price are the lowest, and of course its service life is also the shortest. The write operation of the MLC chip is 3000-10000 times. In contrast to MLC, enterprise-level SSDs usually use SLC technology, and SLC can support 100,000 write operations. Another option is eMLC, which draws on the advantages of MLC and SLC at the same time. It supports 20000-30000 write operations, but the price is lower than SLC. In short, you can get as much as you are willing to spend.

Although manufacturers try to alleviate these problems through write caching and sequential writing, if IT managers can deploy SSDs smartly, they can maximize the service life of SSDs. Here are some good methods:

Step 1: Understand the usage data characteristics of the application.

Most organizations do not understand the characteristics of their application usage data. The common method is to simply deploy the most expensive HDD hard drive, and provide the application oversubscription. This is simple and works well, but it makes capacity management inefficient and brings unnecessary expenses. Most storage vendors provide performance monitoring tools to understand the actual I/O usage. System-level performance characteristics may well divide the proportion of storage occupied by SSDs to meet performance requirements, but they cannot provide any useful information on the service life of SSD devices. The service life is an important part of the total cost of ownership (TCO) of an SSD. Understanding application performance requirements through experimentation is the only way to get sufficient performance with the lowest TCO.

Step 2: Classify applications by read I/O intensity.

After understanding the I/O characteristics requirements of a specific application, the next step is to associate the read-intensive application with the read performance of the SSD. Although these numbers may seem boring, it is very scientific to use them to make decisions. Only a few applications are read-only applications, and there is no doubt that SSD is their best choice. Most applications have read and write requirements, and this read and write ratio is our classification standard. Applications with a high proportion of read operations can benefit from SSDs with fewer side effects.

But here comes the problem, not all read I/O tasks are equal (in fact, when we are talking about read operations, there are more inputs (O) than inputs (I)). I/O is divided into random I/O, sequential I/O and recursive I/O. Performance monitoring tools cannot tell you their categories. Random I/O will not be a problem when the SSD can hold all the data.

Sequential I/O will not benefit from SSD unless all data is stored in SSD. And the application of sequential I/O is usually batch processing, and only run occasionally. If these applications do not have a large number of burst I/O and are not very time-critical, it is not economically cost-effective to place them in an SSD. The deployment of SSDs for recursive read I/O applications is a golden rule, which does not require people to do too much understanding before making this decision.

Step 3: Reasonably allocate SSD storage.

With the above data and forecasts, storage managers need to maximize the use of expensive and limited resources. They need to accurately provide SSD resources with the best performance. Critical applications do not necessarily have high performance requirements, so providing the most expensive storage just because they are important is a misuse of resources.

Step 4: Don't think cost is more important than business value.

Since the IT department has to bear the TCO of storage, it is easy to make the decision whether to deploy or not only by cost. However, if an application requires SSD-level performance, then even if it does not meet the golden rule of SSD lifespan, it should be deployed with SSD. Treating business value as part of TCO will quickly change its calculation method.