- kingfast ssd
- Issue Time
PCIe NVMe SSD is ready to "encroach" the enterprise level storage system
They have installed SSD hard disk all know, want SSD to play all the strength, the need to set the SATA mode in the BIOS to AHCI mode. This can achieve better performance for SSD. At present, the latest storage interface M.2 and SATA-E interfaces are the PCIe channels used, and NVMe is needed for the PCIe interface to perform the highest performance.
Not only for disk enterprise storage is the larger stage of NVMe
In a word, NVMe, like AHCI, is a logical device interface standard. The full name of NVMe is Non-Volatile Memory Express, a nonvolatile memory standard, a specification for SSD using the PCIe channel. The NVMe interface standard has many advantages. First, the NVMe standard is oriented to PCIe SSD. Using native PCIe chanhttp://www.szkingfast.com/products-list.htmnel and CPU direct connection can avoid the delay brought by external communication between SATA and SAS interface, and high performance is the primary advantage of NVMe relative SAS protocol.
Secondly, the early PCIe SSD faces the problem of driver compatibility. Each product has special driver corresponding to different systems. The native PCIe SSD needs loading drive to guide it properly. The emergence of the NVMe standard solves this problem. NVMe SSD can work normally without the manufacturer's corresponding driver. At present, Windows, Linux, Solaris, Unix, VMware and UEFI have joined the NVMe SSD support.
The ease of use brought by high performance and standardization will be revealed in large-scale deployment of PCIe SSD and high load scenarios, such as caching, local storage, distributed storage, ServerSAN and so on. But it needs to be pointed out that today's NVMe has matured. It is not only applicable to a SSD disk, but also a system can also be NVMe, including all flash array, all-flash all-in-one, high-performance SAN and NAS system. This broadens the application of NVMe and is ready to "encroach" the enterprise - level storage system.
The challenge and solution of enterprise level storage system based on NVMe
Dr. Zhang Taile, vice president of Memblaze in the 2017 storage summit introduced to SAS compared with the traditional storage architecture, based on the NVMe enterprise storage architecture directly through the PCIe Switch NVMe SSD CPU is connected, the path is short, because NVMe itself is very good support for multi-core CPU, can make full use of CPU resources, so the performance of the storage system based on NVMe can be very high.
There have been many manufacturers including Pure Storage, are doing all flash array, because of the rapid evolution of distributed software system and iteration, for a long time, based on the distributed x86 server system standard all flash very much, this kind of product by multi copy software level data protection technology to ensure data security and business continuity, high usability considerations system hardware level is lacking.
The total flash array is SAS SSD, and the original NVMe SSD's full memory array is still on the eve of the outbreak. To design a NVMe SSD all flash memory array, we need to disk to the system controller, motherboard, data protection algorithm and even every software and hardware link of the system.
The most important thing for all flash array to do is system level high reliability and high availability. This storage system based on SAS has a lot of reference for all flash array. (the following figure is a storage architecture based on NVMe SSD)
PCIe NVMe SSD preparation
From the above picture, it is not difficult to see that to push NVMe to the enterprise level storage system, we should first do a double port disk to achieve high availability on the disk level. A double port disk is connected to a disk with two servers. A single link failure, such as power loss and short circuit, does not affect the service of the disk to the application. Dual port needs Memblaze and storage arrays and PCIe switches to work together to translate into an effective security force for system high availability. Dual port is the main feature of the SAS disk, which is one of the main reasons that the early PCIe SSD can easily replace the SATA SSD in the system, but it can't replace the SAS SSD.
In addition, high density JBOF design is another key issue in the NVMe era. Zhang Taile said that, compared to JBOD, high density JBOF based on NVMe SSD faces some new challenges. The first is the power supply. A NVMe SSD power consumption of about 25W, about two times as much as the SAS SSD, which means a 60 bit JBOF need to provide power for at least 1500W to the hard disk, and the power of a mainstream 60 bit high density JBOD only 1700W, so JBOF, first of all to the back of the hard disk for power supply the electrical design do to improve. Secondly, after the power is large, the radiator and fan cooling system need more efficiency. If the temperature inside the chassis is too high, it will easily affect the life and stability of SSD.
Complex wiring. The PCIe3.0 X4 interface is commonly used for each U.2 interface, and each Lane needs 4 lines for data transmission. In addition, two clock lines are needed. This means that a 60 - disk JBOF needs to be covered with over 200 lines. This is very high for the design of the backboard and the main board.
For the design of high density JBOF, two schemes are proposed by Memblaze from the design of the U.2 disk. The first is to do low power NVMe SSD. When the power consumption of NVMe SSD is close to the level of SAS SSD, the power of the chassis, backplane and even the design difficulty of the fan will be reduced, even the original JBOD chassis technology can be directly brought to it. It is reported that Memblaze will make a breakthrough on low power NVMe SSD in 2018, when JBOF will have a more easy to use NVMe SSD.
At present, a new generation of Memblaze PBlaze5 PCIe NVMe SSD has the full support of the dual port function, multi host to achieve common access to the device through the PCIe Swich, to avoid the risk of a single host down upper business interruption, improve system availability.