Chapter 8 - 16. Congestion Management in TCP Storage Networks

这篇具有很好参考价值的文章主要介绍了Chapter 8 - 16. Congestion Management in TCP Storage Networks。希望对大家有所帮助。如果存在错误或未考虑完全的地方,请大家不吝赐教,您也可以点击"举报违法"按钮提交疑问。

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. These schemes are called Active Queue Management (AQM). 如前所述,针对在队列中等待的数据包的丢弃或标记方案会极大地影响 TCP 在终端设备上的行为。这些方案被称为主动队列管理(AQM)。

Tail Drop

The tail drop scheme drops newly arriving packets when a queue is full. Essentially, this scheme drops the “tail” of a queue which are the packets that have most recently arrived. Calling tail drop an Active Queue Management mechanism is a misnomer because dropping the packets due to the lack of buffer space is the default behavior and there is no “active” scheme needed for it. 尾部丢弃方案会在队列满时丢弃新到达的数据包。从本质上讲,这种方案会丢弃队列的 "尾部",即最近到达的数据包。将尾部丢弃称为主动队列管理机制是一种误解,因为因缓冲空间不足而丢弃数据包是默认行为,并不需要 "主动 "方案。

Tail drop has an interesting effect on TCP performance. In a case where a single TCP flow arrives when the queue is full, a tail drop results in dropping many packets for that flow, resulting in a significant rate reduction of only that flow. 尾部丢弃对 TCP 性能有一个有趣的影响。在单个 TCP 流量到达时队列已满的情况下,尾部丢弃会丢弃该流量的许多数据包,从而导致仅该流量的速率显著降低。

Another case is when many TCP flows arrive simultaneously. Tail drop results in dropping packets from all of them. As a result, all these flows reduce their rates simultaneously and then increase their rates simultaneously. This results in a cyclic underutilization and over-utilization pattern, called TCP global synchronization. 另一种情况是许多 TCP 流量同时到达。尾部丢弃会导致丢弃所有这些流量的数据包。因此,所有这些流量会同时降低速率,然后又同时提高速率。这就形成了一种周期性的利用不足和利用过度模式,称为 TCP 全局同步。

Random Early Detect (RED)

To avoid TCP global synchronization, packets already queued are dropped (or ECN marked) randomly before the queue is full. This allows only a few TCP connections to reduce their rate. 为避免 TCP 全局同步,已排队的数据包会在队列满之前被随机丢弃(或 ECN 标记)。这样,只有少数 TCP 连接可以降低速率。

Weighted Random Early Detection (WRED)

WRED combines the capabilities of RED along with traffic priority, although the implementation may be different on different platforms. WRED 结合了 RED 的功能和流量优先级,但在不同平台上的实现方式可能不同。

As Figure 8-13 shows, when the egress queue utilizatio文章来源地址https://www.toymoban.com/news/detail-829058.html

到了这里,关于Chapter 8 - 16. Congestion Management in TCP Storage Networks的文章就介绍完了。如果您还想了解更多内容,请在右上角搜索TOY模板网以前的文章或继续浏览下面的相关文章,希望大家以后多多支持TOY模板网!

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处: 如若内容造成侵权/违法违规/事实不符,请点击违法举报进行投诉反馈,一经查实,立即删除!

领支付宝红包 赞助服务器费用

相关文章

  • Chapter 8 - 23. Congestion Management in TCP Storage Networks

    Configuring versus Operating a Shared Storage Network Another important consideration is the operational state of a network. Just because a network is configured for sharing, doesn’t necessarily mean that storage and other traffic flow through it simultaneously. For example, assume you configured a network with 50% bandwidth allocated to iSCSI and NVMe/TCP

    2024年04月09日
    浏览(36)
  • Chapter 8 - 18. Congestion Management in TCP Storage Networks

    Comparison with Lossless Networks Recall that in Fibre Channel fabrics (explained in  Chapter 3 , “ Detecting Congestion in Fibre Channel Fabrics ”) and lossless Ethernet networks (explained in  Chapter 7 ), when the source of congestion is within an end-device, these devices are called slow-drain devices and they are detected using the metrics of the

    2024年03月14日
    浏览(47)
  • Chapter 8 - 25. Congestion Management in TCP Storage Networks

    Modified TCP Implementations Standard TCP with ECN detects the presence of congestion and reacts to it by reducing the congestion window as if packets had been dropped. In data center environments, where the links are of high capacity and end-to-end delay is low, the standard TCP congestion control mechanisms, such as reducing the congestion window by half,

    2024年03月23日
    浏览(33)
  • Chapter 8 - 24. Congestion Management in TCP Storage Networks

    iSCSI and NVMe/TCP with VXLAN Virtual Extensible LAN (VXLAN) extends Layer 2 domains over a Layer 3 data center network. The end devices do not know that their traffic is passing through VXLAN tunnels. Also, the traffic classification based on the DSCP field in the IP header and ECN bits are preserved in VXLAN. Because of these reasons, everything that has b

    2024年03月12日
    浏览(37)
  • Chapter 7 - 14. Congestion Management in Ethernet Storage Networks以太网存储网络的拥塞管理

    PFC Watchdog PFC watchdog works similarly to Pause timeout, but it only drops the traffic in the queue that is unable to transmit continuously for a timeout duration because of receiving PFC Pause frames. PFC 进程看门狗的工作原理与暂停超时类似,但它只会丢弃队列中因收到 PFC 暂停帧而无法在超时时间内连续传输的流量。

    2024年01月22日
    浏览(47)
  • Chapter 7 - 15. Congestion Management in Ethernet Storage Networks以太网存储网络的拥塞管理

    Congestion Notification in Routed Lossless Ethernet Networks End devices and their applications may not be aware of congestion in the network. A culprit device may continue to send (or solicit) more traffic on the network making the severity of congestion worse or increasing its duration. To solve this problem, the network switches can ‘explicitly’ notif

    2024年01月22日
    浏览(51)
  • C Primer Plus 第6版 编程练习 chapter 16

    开发一个包含你需要的预处理定义的头文件。 test.c diceroll.h 两个数的调和平均数这样计算:先得到两数的倒数,然后计算两个倒数的平均值,最后取计算结果的倒数。使用#define指令定义一个宏函数,,执行该运算。编写一个简单的程序测试该宏。 极坐标向量的模(即向量的

    2024年01月19日
    浏览(40)
  • Solidity内存布局介绍 Layout in Memory and Storage

    Solidity reserves four 32-byte slots, with specific byte ranges (inclusive of endpoints) being used as follows: 0x00 - 0x3f (64 bytes): scratch space(暂存空间) for hashing methods  Hash方法的暂存空间. 0x40 - 0x5f (32 bytes): currently allocated memory size (aka. free memory pointer) 0x60 - 0x7f (32 bytes): zero slot Solidity保留了四个32字节的

    2024年02月16日
    浏览(44)
  • 论文笔记-Authenticated Keyword Search in Scalable Hybrid-Storage Blockchains

    混合存储模型: 只有少量meta-data(加密哈希)存在链上,原始数据外包给链下的存储服务商 提出了一个新的ADS 1.首先提出了抑制默克尔倒置(Merkle inv)索引,该索引仅在链上维护部分 ADS 结构,可以使用对数加密证明进行安全更新。 2.提出了一个变色龙倒置(Chameleon inv)索

    2024年01月18日
    浏览(42)
  • 解决RabbitMQ报错Stats in management UI are disabled on this node

    linux 部署 rabbitmq后,打开rabbitmq管理界面。点击channels,会报错 cd /etc/rabbitmq/conf.d/

    2024年02月10日
    浏览(44)

觉得文章有用就打赏一下文章作者

支付宝扫一扫打赏

博客赞助

微信扫一扫打赏

请作者喝杯咖啡吧~博客赞助

支付宝扫一扫领取红包,优惠每天领

二维码1

领取红包

二维码2

领红包