Shulou(Shulou.com)06/01 Report--

This article shows you how to use multi-stream writing technology to improve the performance and life of SSD, the content is concise and easy to understand, can definitely make your eyes bright, through the detailed introduction of this article, I hope you can get something.

Learn the meeting materials of FMS16 today. There is a Samsung multi-streaming technology, Samsung is recommending it as an international standard, which I think is more interesting, so let's talk about it today.

First of all, we know that some of the features of NAND flash memory are different from hard drives: it has three operations, read, programming (which is actually writing), and erasing. The hard disk supports overwriting, so there is no need to erase. Flash read and write units are page, but erase units are block (that is, multiple page). Since writes cannot be overwritten, there is an extra act of garbage collection, that is, copy the useful page data somewhere else, and then erase the block before you can rewrite the data. This brings additional write magnification, and data movement must be locked during garbage collection, so in addition to affecting write performance, read performance is also affected.

But another difference between flash memory and hard disk is the problem of lifespan. That is, the number of page programming is limited (in fact, hard drives also have a lifetime, but relative to flash memory, everyone thinks that there is no such problem in its life cycle). According to Samsung statistics, the write lifetime of different flash memories is declining, which should not be a technical problem, but the result of the pursuit of higher density. However, with the advent of 3D-NAND technology, there should be some improvement, because 3D has more space and less interference with each other. Generally speaking, the life span of 3D TLC NAND is similar to that of 2D NAND MLC.

To this end, Samsung proposed a multi-stream writing technology, that is, the application divides the data into multiple data streams according to different update frequencies, and when SSD writes, it recognizes these data streams and writes the data streams with the same characteristics together (continuous block).

The advantage of this data layout is that garbage collection is more efficient, because the frequency of garbage collection varies according to the characteristics of the data stream, and the block occupied by some infrequent data need not be recycled. Generally speaking, the recycled block block data are invalid and can be erased directly to avoid data replication.

By avoiding more page data replication and reducing write magnification, it is equivalent to prolonging the write life of the system.

Samsung tested Cassandra, an open source No-SQL database, before and after the multi-streaming patch, and found that it could increase write life by two times, improve performance by 45%, and reduce read latency by 50%. It's still a very good result.

Samsung is currently promoting this multi-streaming technology as a standard. At present, the SCSI/SAS interface has been completed, and NVMe is doing a final review.

I downloaded the standard to have a look, it is still a draft, not a formal standard.

In fact, the standard mainly defines the format of initiator and target communication for SCSI. Let the operating system tell SSD the identity of the data stream through the SCSI command when writing the disk. If it is NVMe, because the SCSI layer is by pass, you need to add the command of NVMe, so it is a different standard. In addition, after standardization, we need operating system support, as well as various application support. Because only the application finally knows the characteristics of its data, it's best to let it mark it.

In fact, AFA (all flash array) can also take advantage of this idea. For example, OceanStor Dorado V3, which Huawei is about to release at the HC conference in Shanghai, is said to adopt a global FTL, using hot and cold data separation to improve the efficiency of garbage collection.

This hot and cold data separation is actually the same as multi-stream writing. The difference is that the multi-stream writing proposed by Samsung needs to be supported by an application system or operating system, and only for a single SSD. The cold and hot separation of Huawei is implemented within the storage and is at the system level. Huawei can do this, mainly by using a self-developed HSSD disk, which allows the array software to work with SSD's fireware.

However, if Samsung's multi-stream writing becomes a standard, Huawei's Dorado V3 will also benefit from it. Because the application tags the written data directly and sends it to the array, it is more accurate and easy for the array to judge hot and cold data, and it will be more efficient. Moreover, with these data flow tags, the Cache scheduling algorithm of the array will be more efficient, and you will save the hottest data to Cache as much as possible to improve the hit rate of Cache.

Flash memory technology brings many design changes to the whole storage system, and the intelligent layout of data has a great impact on the performance and life of flash memory. Future AFA systems will be smarter and more application-aware.

The above content is how to use multi-stream writing technology to improve the performance and life of SSD. Have you learned the knowledge or skills? If you want to learn more skills or enrich your knowledge reserve, you are welcome to follow the industry information channel.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.