Introduction: SSD solid-state drive is a storage device composed of a control unit and a solid-state storage unit (DRAM or FLASH chip). Currently, it supports SATA, FC, and SAS interfaces and can replace ordinary mechanical hard drives.

1. What is SSD

SSD, also known as Solid State Disk or Solid State Drive, is a storage device composed of a control unit and a solid-state storage unit (DRAM or FLASH chip). It currently supports SATA, FC, and SAS interfaces and can replace ordinary mechanical hard drives.

SSD disks are divided into SLC SSD and MLC SSD based on capacity, reliability, and performance requirements. SLC is mainly used in high feasibility environments such as servers, workstations, and storage, while MLC is mainly used in environments with low reliability requirements for desktop PCs.

The comparison of the internal structure diagram between SSD drive and traditional mechanical hard drive is as follows:

Internal structure diagram of SSD disk and mechanical hard disk

Compared to traditional mechanical hard drives, SSD hard drives have the following advantages:

1. Fast data access speed. Due to the use of flash memory chips, there are no mechanical components inside the SSD solid-state memory. SSD performs well compared to traditional hard drives in terms of intensive IO read and write and low latency.

2. Low failure rate. One of the characteristics of SSDs is shock resistance and impact resistance, as they all use flash memory chips. Therefore, there are no mechanical components inside the SSD solid-state memory, which will not affect normal use even during high-speed movement or accompanied by flipping and tilting. Moreover, the possibility of data loss in the event of accidental dropping or collision with hard objects of the laptop computer is minimized.

3. Green and environmentally friendly. Solid state memory is silent during operation (due to the lack of mechanical motors and fans, the noise level during operation is 0 decibels), with low heat generation and fast heat dissipation.

4. Lightweight. Solid state memory is lighter in weight, making it easier to carry on mobile devices.

At present, the annual growth rate of HDD mechanical hard disk capacity is relatively slow, and the read and write speed has basically not increased in the past 10 years. However, the processor speed has increased exponentially according to Moore's Law. HDD hard disk storage media has become the bottleneck of computing systems. With the development of the semiconductor industry, the integration of Flash is becoming higher and the cost is continuously decreasing. With the adoption of multi-channel parallel technology, SSD disks are becoming more and more widely used in high-end servers, storage, and portable computers due to their fast read and write speed, green power saving, and high reliability without mechanical components.

2. Classification of SSDs

There are currently two mainstream SSD products, SLC and MLC. The main difference between the two products lies in the number of erasures and performance.

The full name of SLC is Single Level Cell, which refers to a Block (the basic storage unit of Flash, also known as Cell) with only two types of charge values. Different high and low charge values indicate 0 or 1, because only a set of high and low voltages is needed to distinguish 0 or 1 signals. Therefore, the large driving voltage of SLC can be very low. Traditional dual voltage cards or low voltage versions of cards definitely use SLC type NAND Flash chips.

SLC has a simple structure and a small range of voltage changes when writing data, resulting in a longer lifespan. Traditional SLC Flash can withstand 100000 reads and writes, so the probability of bad blocks is low. Because the storage structure is very simple and can be driven by a set of voltages, its speed performance is better. Currently, all ultra high speed cards use SLC type Flash chips.

MLC (Multi Level Cell) is a technology that fully utilizes blocks. It is driven by higher voltage and records two sets of bit information (00, 01, 11, 10) in a block using different levels of voltage. This can double the theoretical recording density of the original SLC. However, in addition to the theoretical double recording space for the same process and same crystal area, MLC has some inherent drawbacks. For example, if the voltage range is smaller, Flash requires more CRC verification space, which will occupy approximately 10% of the space in the Block. Therefore, in actual use, the capacity of MLC with the same process and same crystal area is less than twice that of SLC.

Due to more frequent voltage changes, MLC technology's Flash is far inferior to SLC in terms of lifespan, with an official erasable count of only 10000 times, which is a fatal drawback of MLC.

The Flash of MLC technology also has a disadvantage, as its read and write speed is not as fast as SLC. One block stores two sets of bit data, which naturally requires longer time. There are also factors to consider, such as voltage control and CRC write method.

3. SSD Technology Analysis

NAND FLASH Loss Leveling Algorithm (Wear Leveling)

Because the number of write operations in flash memory is limited, if 100000 write operations are performed on certain units, the reliability of subsequent write operations for these units cannot be guaranteed, and some units may fail. For example, some management data system logs are frequently rewritten, while some static file data is almost never rewritten. If not controlled, some blocks will be damaged prematurely due to repeated erasures, while others have not been rewritten, thereby affecting the service life of the SSD disk. To solve this problem, write operations are evenly distributed across all flash units to achieve overall balance and avoid individual unit failures. Wear Leveling algorithm is widely used to solve this problem. Algorithm.

Wear Leveling provides a block mapping mechanism that disperses write loss across different blocks, without causing certain blocks to be written out first and causing the entire SSD disk to fail. Instead, it replaces blocks that fail before their expected lifespan with reserved blocks. This algorithm ensures that the lifespan of the entire device is on the same level as the lifespan of Flash.

Generally, a page based file storage algorithm is used to implement the Wear Leveling algorithm, and there is no one-to-one correspondence between the physical and logical addresses of flash memory. When a solid-state drive receives a data write request, it does not follow the usual order of writing, but instead finds the units with fewer writes to write to. Therefore, when dynamically allocating physical blocks for writing data, corresponding priorities will be assigned based on the usage of each block, thereby balancing the lifespan of the entire storage unit.

SSD multi-channel concurrent technology

Due to the current 8-bit data bus of Nand Flash, it can provide a read speed of 25MB/s and a write speed of 3MB/s. Obviously, if SSDs provide this speed, it is unacceptable for users. Currently, SSD controllers generally use multiple channels to operate multiple Flash chips in parallel, similar to RAID0, which greatly improves the read and write speed, as shown in the following figure:

Error correction management and mapping mechanism

The SSD controller monitors and verifies the data of reading and writing Flash to ensure the success of data operations, and has built-in error correction redundant data to prevent the loss of individual unit data based on flash devices from causing the loss of the entire data block.

The mapping mechanism of SSD generally adopts a page based file storage algorithm, and there is no one-to-one correspondence between the physical and logical addresses of flash memory. When a solid-state drive receives a data write request, it does not follow the usual order of writing, but instead finds a convenient and fast location for writing to improve the write speed.

Usually, the detection work verifies the success of write and erase operations through on-chip status registers and specific hardware components, and the verification work is completed by comparing the written data with user data through readback. When a write operation fails, the system does not simply report the error to the user. Instead, it uses a block remapping mechanism to allocate an additional free block to re execute the write operation. During the design of the SSD, a portion of Flash has been reserved for temporary storage of data and replacement of bad blocks. In addition, the Flash chip has a limit on the number of erasures and reads, so even if some blocks are written bad, it will not affect the service life of the entire SSD. Therefore, SSDs prevent errors from occurring through automatic verification, and individual unit read errors can be recovered through error correction to ensure data integrity. The mapping mechanism between logical addresses and physical addresses solves the problem of writing physical block errors, and the mapping mechanism between logical addresses and physical addresses is also needed in loss balancing algorithms to achieve this.

4. SSD application environment

The characteristics of SSD disks compared to ordinary mechanical hard drives are the same sequential and random data reading performance, with a response time of less than 1 millisecond, no mechanical components, and high reliability. After careful research, analysis, and testing of SSD disks from multiple manufacturers, H3C has chosen SLC SSD disks with high reliability and performance to work well with 10 Gigabit storage. Each disk can achieve a read performance of 250MB/s and a write performance of 170MB/s. With 4 KB reads, each disk can reach 35KIOPS, and with 4 KB writes, each disk can reach 3.3KIOPS.

SSD disks have high performance and reliability, but their prices are relatively high and their capacity is relatively small. This determines that SSDs are more suitable for applications with high random read ratios and high response time requirements, mainly including the following:

Database environment. The data disk of a database is usually a random read and write of small data blocks, with high response time requirements. The majority of them have a capacity of less than 2TB. RAID5 composed of SSD disks can provide high IOPS while maximizing its capacity.

Shared Data Disk for High Performance Cluster Systems. In a cluster system, a large amount of data needs to be shared, and the data needs to be frequently accessed by various nodes, requiring fast response speed. In this system, by using SSD disks, the waiting time for each node to access the shared disk can be greatly reduced, and its performance can be improved.