代码编织梦想

chapter 8 -爱代码爱编程

Queue Depth Monitoring and Microburst Detection Queue depth monitoring and microburst detection capture the events that may cause congestion at a lower granularity but are unnoti

chapter 8 -爱代码爱编程

Bit Errors in Lossy Ethernet Networks with TCP Transport As explained in Chapter 1, when a bit stream is exchanged over a network, some bits may be altered resulting in bit error

chapter 8 -爱代码爱编程

TCP Flow Monitoring versus I/O Flow Monitoring TCP flow monitoring shouldn’t be confused with I/O flow monitoring because of the following reasons: TCP 流量监控不应与 I/O 流量监控混淆,原因如下:

chapter 8 -爱代码爱编程

Switch Buffer Management Recall that during network congestion, a TCP sender relies on the following events for reducing its transmission rate. 回想一下,在网络拥塞期间,TCP 发送端依靠以下事件来降低传输速率。

chapter 8 -爱代码爱编程

Comparison with Lossless Networks As Chapter 2 explains, Fibre Channel defines a maximum frame size of 2148 bytes. Users do not have to change the MTU value, and hence, it is nev

chapter 8 -爱代码爱编程

Comparison with RoCEv2 Networks RoCEv2 Congestion Management (explained in Chapter 7) uses the same mechanism as ECN in TCP/IP networks. Compare Figure 8-103 with Figure 7-17 in 

chapter 8 -爱代码爱编程

Congestion Avoidance When the sender’s upper layer desires to send more data, TCP throughput increases assuming that the receiver-window is not a limiting factor. But this increa

chapter 8 -爱代码爱编程

Congestion Notification in TCP Storage Networks Congestion Notification in TCP storage networks is achieved by Explicit Congestion Notification (ECN), which is standardized by RF

chapter 8 -爱代码爱编程

Active Queue Management As previously mentioned, dropping or marking schemes for packets that are waiting in a queue can significantly influence TCP’s behavior on the end devices

原语,原子,线程安全-爱代码爱编程

原子操作和原语是计算机科学中常见的概念,通常用于多线程或多进程环境中,以确保数据的一致性和同步。 原子操作(Atomic Operations) 原子操作是不可再分的操作,在执行完毕之前不会被线程调度系统中断的操作。从外部看,一个原子操作要么是完全执行完成的,要么是未执行的,没有中间状态。这种特性使原子操作成为实现同步机制如互斥锁(mutexes)和信

chapter 8 -爱代码爱编程

Selective Acknowledgment (SACK) Some TCP implementations use Selective Acknowledgment (SACK), which increases performance during packet drops by selectively acknowledging only th

chapter 8 -爱代码爱编程

Congestion Due to Over-utilization of the Target Link The same problem may happen on a target/storage link when multiple hosts initiate large-size write I/O operations to the sam

chapter 8 -爱代码爱编程

NVMe/TCP I/O Operations Figure 8-7 shows NVMe/TCP Read I/O Operation. A host (initiator) initiates a read I/O operation over a TCP connection to the target, which is called a con

linux -爱代码爱编程

per-CPU变量 per-CPU变量是一种存在与每个CPU本地的变量,对于每一种per-CPU变量,每个CPU在本地都有一份它的副本。 per-CPU变量的优点 多处理器系统(smp)中无需考虑与其他处理器的竞争问题

chapter 7 -爱代码爱编程

Layer 2 Priority Flow Control As Figure 7-4 shows, at layer 2 of the OSI model, traffic is assigned to different VLANs by adding an IEEE 802.1Q VLAN header. This VLAN header cont

chapter 7 -爱代码爱编程

Pause Threshold for Long Distance Links长途链路的暂停阈值 This section uses the following basic concepts: 本节使用以下基本概念:  Bit Time (BT): It is the time taken to transmit one bit. It is the

chapter 7 -爱代码爱编程

Stomped CRC Counters Stomped CRC counters help in finding the location of bit errors in a network that uses cut-through switches. More precisely, these counters help in finding w

chapter 7 -爱代码爱编程

Case Study 2 — Converged Versus Dedicated Storage Network This case study is similar to Case Study 1. The difference is that the server is not running into a high CPU utilization

chapter 7 -爱代码爱编程

Storage I/O Performance Monitoring Traffic in a storage network is the direct result of an application initiating a read or write I/O operation. Because of this reason, network t

chapter 7 -爱代码爱编程

Bit Rate Differences between FC and FCoE Differences in the bit rates must be carefully accounted for when a switch transfers traffic between FC and FCoE ports. As explained in C