NVMe

翻译一下，纯粹是为了记住这些特性：

NVMe provides the following benefits:
● Ultra-low latency 非常低的延迟
● Very high throughput 非常高的吞吐量
● Low power architecture, delivering lower power consumption and resulting in a lower Total Cost of Ownership and reduced carbon footprint 低功耗设计，提供更低的TCO以及碳消耗（扯得有点远，不过有道理，谁让节能减排是主流呢）
● Reduced gate count 减少了门数（什么意思?难道是数据流层数减少了？）
● Cost reduction and shorter time to market through use of a standardized interface 降低成本，且更快地切入市场
● Reliable performance across multiple cores enabling quick access to critical data 性能可靠，支持多核模式，保证更快地获取关键信息
● An optimized register interface and command set that reduces CPU utilization resulting in higher performance and lower power 优化了寄存器接口和指令集设计，降低了CPU处理的复杂度，显著提高性能，降低功耗
● Scalability with headroom for current and future NVM performance 为现在和将来提供了可拓展空间（狡兔三窟）
● End-to-end data protection capabilities and support for standard security protocols, such as Trusted Computing Group 端到端数据保护，支持标准安全协议，如TCG (背靠大树)
● Seamless integration into multiple operating system environments with standard open driver interfaces 使用标准开源的驱动接口，可以无缝整合到多操作系统的环境中

NVMe - Architected for Performance 为性能而生
• No practical limit on number of outstanding requests 对请求数量基本没有限制
 Up to 64K I/O queues each with up to 64K entries 高达64K个Queue,每个Queue高达64K个Command
 32-bit controller unique command identifier (16-bit Queue ID + 16 Command ID) allows up to 232 outstanding commands ？
• Supports many-core processors without locking 支持多核，无锁设计
 Each processor may be configured with its own submission/completion queues and MSI-X interrupt 每个核可以被单独配置，拥有自己的submission和completion queues以及MSI-X中断
• At most one doorbell write to issue a command 一个Doorbell就可以触发命令执行
 Multiple commands may be issued with a single doorbell write 多个命令使用一个Doorbell就可以触发执行
• Streamlined NVM command set avoids burdening controller with legacy command support requirements 流水线式的设计，降低控制器的压力
• Fixed size 64B commands and 16B completions enable fast and efficient command decode and execution 命令的大小固定，解码执行更快更高效
 Commands contain 2 PRPs allowing 4KB or 8KB reads and writes be processed without fetching any additional information (e.g., scatter/gather list) 命令包含了2个PRP，支持4KB和8KB的连续顺畅读写
• PRP based scatter/gather list allow efficient out-of-order data delivery  PRP支持高效的乱序数据传输

NVMe Standardized Features
• Logical block data and metadata 逻辑块数据和元数据
• End-to-end data protection (T10 DIF and DIX compatible) 端到端保护
• Security (Trusted Computing Group collaboration) 安全
• Submission queue arbitration and QoS 服务质量保证
• Firmware update and activation 固件更新、激活
• Dynamic power management 动态电源管理
• Robust error reporting 高大威猛的报错机制
• Interrupt coalescing configuration/control 中断联合配置、控制
• Capability discovery and configuration 容量检测和配置

正所谓

早生晚生不重要，出身高贵乃正道

倘若不服你来辨，性能好到你出汗