Traffic Control Algorithms for Integrated-Services Packet-Switched Networks

在这项研究中，我们探索交通控制算法， 综合业务网。这些网络承载着 不同的业务类型，例如计算机数据、语音、视频、文本， 和图像。 在这样的网络性能目标和数据包 从一个数据流到另一个数据流， 另 两个网络都必须设计有如此大的容量 每一个可以想象的性能目标集都将得到满足， 或者需要区分分组和分组的算法 流基于这些特征并调整延迟 分布、损失概率和阻塞概率 相应地 我们计划开发一套完整的交通系统 控制算法部分基于一种称为 优先令牌银行。 接纳控制算法接纳或 阻止整个分组流，其中允许分组流 意味着保证其绩效目标得到实现。在 对于每个队列，调度算法命令传输 排队的数据包和丢弃算法选择数据包 当缓冲区溢出发生时丢弃。 两种算法 有效地调整延迟分布和损失概率， 在给定负载下优化性能， 应用，或者，等同地，实现给定的水平， 在更高负载下的性能。我们还探索基于使用的收费 对于基于分组到达过程的网络服务， 已接纳的数据包流以及相应的性能 实现（并可能得到保证）。这些费用反映了 网络资源消耗的程度。警务和 监测也得到了解决。算法通过以下方式进行评估 分析和广泛的模拟，通过比较延迟，损失 概率、阻塞概率和其他相关的 度量，与那些用已知算法实现。

In this research, we explore traffic control algorithms for integrated- services networks. which are networks that carry diverse traffic types such as computer data, voice, video, text, and images. ln such a network performance objectives and packet arrival processes vary tremendously from one data stream to another. Either networks must be designed with so much capacity that every conceivable set of performance objectives will be met, or algorithms are needed that differentiate packets and packet streams based on these characteristics and adjust delay distributions, loss probabilities, and blocking probabilities accordingly. We plan to develop an integrated set of traffic control algorithms based in part on a novel mechanism called the priority token bank. An admission control algorithm admits or blocks entire packet streams, where admitting a packet stream means guaranteeing its performance objectives will be met. In each queue, a scheduling algorithm orders the transmission of queued packets and a dropping algorithm selects packets to be discarded when buffer overflow occurs. Both algorithms effectively adjust delay distributions and loss probabilities to optimize performance at a given load as perceived by the applications, or, equivalently, to achieve a given level of performance at a higher load. We also explore usage-based charges for network services based on the packet arrival processes of admitted packet streams and on the corresponding performance achieved (and possibly guaranteed). These charges reflect the extent to which network resources are consumed. Policing and monitoring are also addressed. Algorithms are evaluated through analysis and extensive simulation by comparing delays, loss probabilities, blocking probabilities, and other relevant metrics, with those achieved with known algorithms.