With the large number of applications of solid-state drives in the market, the applications and technologies such as AI, artificial intelligence, and high-specification video in the era of big data have increasingly high storage requirements. The number of NAND layers of SSDs is becoming more and more dense, the master control performance is also increasing year by year, and the technologies and specifications related to SSDs are also updated.
Among them, SATA took the lead in becoming the main way to connect SSDs in personal computers. However, since SATA was designed as a serial interface and adopted the AHCI specification, there was only 1 command queue with a depth of 32. With the passage of time, it becomes increasingly difficult to meet the increasing speed of SSDs. As a result, high-end SSDs began to use the PCIe bus to provide higher performance, but because there was no unified standard, they needed to use non-standard specification interfaces.
NVMe's technical work began in 2009, with NVM Express working groups from more than 90 companies around the world developing specifications. Because it has 65535 command queues, each queue can reach 65536 commands, and it also makes full use of MSI's 2048 interrupt vector advantage, greatly reducing latency. NVMe's performance and standardization improvements have become the norms that many companies respect.
In 2011, NVMe 1.0 version was officially launched, with a maximum bandwidth of 1200MB/s;
In 2012, IDT released the first commercial NVMe chipsets;
In 2013, Samsung launched the first NVMe SSD;
In 2014, NVMe officially incorporated the Group, which consists of more than 65 companies from the industry to jointly maintain and promote the implementation of the NVM Express specification;
In 2017, the launch of NVMe 1.3 was widely used on PCIe Gen3 SSDs;
In 2019, NVMe 1.4 was launched, becoming standard on PCIe Gen4 SSDs;
In June 2021, NVMe 2.0 was launched, and by January 2022, NVMe 2.0b was launched for further improvement. The reconfigured NVMe 2.0 specification enables faster and easier development of NVMe solutions, adding support for mechanical hard drives and adding extensibility command sets such as partition namespaces (ZNS) and key-value (KV) and other new features.
After several updates, ZNS in the NVMe 2.0 specification now supports partitioning based on data usage frequency and storing them sequentially in separate areas within the SSD, thereby reducing rewriting and rearrangement of stored data, reducing the write amplification factor (WAF) of the hard disk, and ultimately achieving an increase in the performance and longevity of the overall drive.
ZNS also reduces the need for DRAM cache capacity on SSDs. On traditional SSDs, DRAM cache needs to be configured in a 1GB:1MB ratio to ensure that the SSD maintains excellent storage speeds. On ZNS SSDs, the capacity of each mapped partition rises from 4KB to tens or hundreds of MB, saving more DRAM cache capacity for larger enterprise-class HDDs.
ZNS is expected to reduce drAM cache capacity requirements by 8 times and OP over-provisioning requirements by about 10 times. On this basis, it not only reduces the cost of DRAM cache configuration, but also increases the capacity of the hard disk that users can use, not only for consumer-grade SSDs, but also for larger capacity enterprise-class SSDs.
The KV (key-value) instruction set is another important feature of NVMe 2.0, and the KV command set will allow applications to communicate with the hard disk controller using KV pairs instead of through block addresses. To eliminate unnecessary translation tables between keys and logical blocks, reduce CPU computing load and thus reduce overall overhead.
Endurance Group Management is also an important feature of the NVMe 2.0 specification, through this new storage management mechanism, SSD configuration can be flexibly and dynamically adjusted for dynamic capacity management and hybrid NAND operation. To improve access to SSDs and achieve better control over storage devices as a whole.
The realization of the above functional points is inseparable from the main control chip of the solid state drive. According to the ZNS partition namespace, Huirong Technology's exclusive hardware and firmware technology can arrange the workload on different microprocessors according to different application requirements, and at the same time isolate the main control chip while realizing zone partitioning in NAND Flash, so as to achieve more flexible control regulation and higher performance optimization.
Under the many advantages and improvements, NVMe will bring longer service life, more flexible management mechanism and lower cost to enterprise SSDs, and will become the next important technology node trend of enterprise SSDs. As the world's leading manufacturer of flash memory master control chips, Huirong Technology has always kept up with the industry's dynamics and actively responded to innovative technologies and specifications, and NVMe 2.0 products and solutions are ready to provide enterprise-class storage solutions with higher performance and more BOM advantages for data centers in the era of big data.