Analysis of SSD disk array

Focus on SSDs has been around for some time, the previous few years SSD capacity is small, more importantly, its limited life and expensive prices so that SSD can only be used as a storage area of the accelerated media exist, if anyone who want to use SSD as a disk array, the system is a bit too crazy. As a result, in previous years, both academia and industry have been using SSD as an accelerated medium, for example, as a cache, I have developed a system that uses SSD as a cache for disk arrays. That year's start-up company Fusion-io developed the PCI-E bus interface of the SSD disk by the storage company's favorite, until today, many of the key business acceleration can be used Fusion-io high-performance SSD disk. The company is now listed.

Technology development today, SSD regardless of performance or in the capacity has made great progress, the application of high-performance storage needs more and more strong, the idea of using SSD to build disk array is gradually reliable. To this end, there are a number of Silicon Valley for SSD-oriented entrepreneurial companies, which is more typical of Pure-storage, the company launched the SSD disk is very unique.

As a technology developer, we can not help but ask: SSD disk array and disk array in the technical aspects of the special difference? Can the traditional disk array technology be used directly to the SSD disk array?

The implementation of storage technology is always around a bit, that is, the storage media read and write behavior. Regardless of the file system or the IO dispatch layer of the operating system, all technical considerations revolve around the read and write behavior of the storage media, so many of the current policies in the operating system are for disk media features. What are the characteristics of the disk? The biggest feature of the disk is the existence of seek time, if the disk is very poor performance in random Read and write mode. To this end, the operating system will use Pre-fetch to reduce disk read and write times, the Io-schedule layer using IO aggregation sort to reduce disk jitter.

SSD is not a mechanical disk, its use of semiconductor storage media, therefore, do not have the characteristics of traditional disk, so we use SSD, the software level of the technology adopted will be changed, which we can glimpse from the Linux open source project. For example, Linux adds the trim mechanism for SSD and requires modifications to the Scsi,io-schedule,md-raid and file systems to enhance SSD performance. In fact, because of the changes in the read and write features of SSD, the technology used in designing SSD disk array is necessarily different from the traditional disk array. Here, I am very convinced that the following major technologies will be very beneficial to the development of SSD disk array technology:

1,deduplication. As a result of disk random read and write performance problems, duplication of data deletion and disk system is basically missed. If it is a backup system, it will be another matter, because the data of the backup system is very strong order, so it can compensate the disk random read and write performance by locality high cache. The random read-write performance of SSDs is high, and the latest test data from Samsung show that SSD can achieve 80K of random read and write ioPS. Therefore, the deduplication technology can be applied to SSD disk array completely. The disk array with duplicate data deletion can achieve the purpose of "calculating IO", and the system performance can be improved with the improvement of CPU performance; On the other hand, deduplication can make up for the small size of SSD. In many applications, high-performance deduplication can do 20:1 of the data compression rate, which is a very optimistic number.

2,data-log. SSD a very good performance is random read and write performance, and a headache is the problem of writing amplification, lowercase for SSD in the firmware is a negative operation. and log technology can make use of SSD's superior random read-write performance to compensate SSD's lowercase problem. Log technology is also used on traditional disk, the well-known DCD technology is the use of disk excellent order performance to compensate for inefficient random write performance. However, because of the poor random read performance of disk, disk-oriented DCD technology can only be used in write applications, and read performance is very poor, can not be applied. SSD-oriented log technology does not exist this problem, excellent randomness can be fully qualified for the application of the demand.

I would like to take a deep look at the read and write features of SSD, using the technology that conforms to the features of SSD, can make SSD disk array achieve the best performance and service life.

This column more highlights: http://www.bianceng.cnhttp://www.bianceng.cn/database/storage/