How is data protected on SSDs?

from Store Old Wu Blog--The way of storage

The development of Flash technology has reduced its cost and increased capacity, and more and more enterprise-class users are also considering how to make Flash SSD technology is applied to your environment to address storage problems such as performance, power consumption, and volume. Flash technology has been widely used in the Internet in previous years, mainly because the internet needs to face a lot of access IO pressure, traditional disk storage system can not provide high IOPS capacity, so The high performance of Flash can only be used to solve the problem of Internet development. As a result, today's flash presence in the Internet domain has been widely used and has become standard.

in the process of flash applications,PCIeThe application of the interface to storage has been a great success. Especially in recent years in line withNVMeinterface Standard forSSDlaunched, it is accelerating the flashSSDthe application. Last yearIntellaunched a complianceNVMeStandard of2.5inchSSDthis year, the domesticMemblazealso launched in line withNVMe1.1Standard of2.5inchSSDdisks and cards. Can say this2.5inchSSDthe launch of the disk clears the barriers to the presence of enterprise-class applications for Flash. 2.5inch disks have hot-swappable features and feature enterprise-class dual-port properties. In theNVMeWhile the drive is on the ground, server vendors are strongly supportingNVMe, from last year to this year, many server manufacturers launched products will supportNVMeinterfaces, common to some2Uthe server will support4a few or so.NVMedisk. Ultra-Micro launched1UServer can supportTenFull-dialNVMe SSD. Standard ofNVMedisk has, supportNVMethe server has, then now the problem comes, how in these servers will so manyNVMeWhat about the plate?

nvme SSD nvme SSD It's no difficulty to run directly. However, we need to know that the traditional storage software is designed for disk media, he considers the problem of disk media, such as poor random access, sequential access ability is good. For these characteristics of disk media, traditional storage software has been developed. Compared to disk, nvme SSD is completely different, He has no problems with disk storage, but he has his own unique problem. Therefore, if the traditional storage software is directly architected in the nvme SSD nvme server platform, we can not directly, Simple use of existing software for disk storage to solve nvme SSD Data reliability, volume management and other thorny issues, need to be a further alternative.

in the case of data protection, NVMe SSD what should be done on the platform? What are the problems with traditional solutions? Where are the technical directions for the future? Store old Wu to give you one by one way to come.

in the era of disk storage, it is common to useRAIDTechnology for data protection, multiple copies of the Internet can be understood asRAID1and its extension. RAIDtechnology can often be divided into softRAIDand HardRAIDtwo major categories. HardRAIDis achieved by means of hardware. In theRAIDat the beginning of the development of technology, we adoptFPGAWay to achieveRAIDController, that yearXinlixLaunch ofFPGAintegrated in theXORUnit to support developmentRAIDThe controller is also directly developed by the manufacturer.ASICto order, speed upRAIDperformance. The most common use of course isIntelof theIOPprocessor, and nowLSI ROCwaitSOCtechnology. From which we can realize that the real life of the product architecture is simple, and software isITthe core of the product.

650) this.width=650; "title=" 1.jpg "src=" Http://s3.51cto.com/wyfs02/M01/70/54/wKiom1W2Xi-iyDmbAACllXLqcmc566.jpg " alt= "Wkiom1w2xi-iydmbaacllxlqcmc566.jpg"/>

HardRAIDis a based onPci/pciebus interface card, on-card expansionSATAinterface, and inSOCRun inRAIDfirmware program. The whole hardRAIDThe core technology is the firmware program andSOCchip. HardRAIDis a standard for a lot of storage, he has a very strong universality. In addition to hardRAIDtechnology, many manufacturers will develop softRAID, such asEMCand other storage manufacturers will have their own softRAIDtechnology. SoftRAIDis run on the host side of the data protection software, through the softRAIDcan be basically achieved with the hardRAIDthe same data protection, but no hardRAIDso strong in universality. Ten years ago,CPUperformance is far less than now, so the use of softRAIDtechnology, causingCPUparticipate in the coding and decoding of data, which can affect performance. So, in a long time ago hardRAIDare representative of high performance. But, now ,Intel CPUhas been very powerful for handling the softRAIDThis kind of data codec is a cinch, not to mentionAVX,SSEsupport for accelerated instruction sets. As a result, today's softRAIDperformance is not inferior, even beyond the hardRAIDthe performance.

whether it's soft RAID or a hard RAID technology in the disk storage era has been very good, in addition to and RAID schema-related data refactoring itself. In normal use,theRAID itself is acceptable, because the main performance bottleneck is the disk itself. So, from a research point of view, the vast amount of work in that era is optimizing IO Patternto meet the needs of disk storage.

times into flash storage,SSDIf you directly access the traditionalRAIDsystem, we do not consider the traditionalRAIDtheSSDitself caused by a number of problems, from a performance point of view, the traditionalRAIDhave become the bottleneck of the system. An ordinarySSDdisk, whether it isSATAor isNVMeinterface, its random read performance is easily achieved byTenmillionIOPS, the performance is easily achieved after the multi-disk aggregation -million or more. Under this kind of performance pressure, a common hardRAIDHow will this performance be played out? For aNVMe SSDThe single-disc performance of the disc is achieved -millionIOPSabove, if using5BlockNVMe SSDBuild aRAID, then how to play the random read performance to -million or more? With HardRAIDIs it realistic? If this performance is to be achieved, then we need to redesignRAID SOCcontroller requires at least one multicore high-performance processor to implement thisSOC. Suppose a nucleus can handle -millionIOPS, then thisSOCneed at least6such a nucleus. This does not consider the software-level introduction of performance loss, you knowRAIDThe coupling of software system is very strong, it is very difficult to exert multi-core performance to the extreme. In addition, to achieve -millionIOPSperformance, bandwidth reached12gb/sabove, apcie3.0x16just met the requirements. So, from a hardware point of view, the short-term implementation of such a commercial hardRAIDis unrealistic. At present in the market, many people also adopt the hardRAIDfor a relatively low performanceSATA SSDfor data protection.

for NVMe SSD , the use of hard RAID in a hopeless situation, only the data protection software can be considered to enhance the reliability of the system data. This is actually in line with the objective law,the development trend ofCPU is multicore,the operation ofNVMe SSD is mainly IO - intensive operation, Very consistent with multi-core development ideas. With multi-core technology , IO requests can be processed concurrently , enabling the performance benefits of flash memory to be exploited. So, storage Lao Wu believes that under the Intel X86 Server platform, the full use of multicore technology, the characteristics of flash memory for the NVMe SSD Building data protection software system is the trend of storage technology development in the future.

This article is from the "Save the Way" blog, make sure to keep this source http://alanwu.blog.51cto.com/3652632/1678956

How is data protected on SSDs?