论文翻译

Poster: ACP: Age Control Protocol for Minimizing Age of Information over the Internet

海报：ACP：最小化信息在互联网存在时间的年龄控制协议。

ABSTRACT Real-time monitoring is characterized by a source repeatedly sending updates over the Internet to a monitor, which desires the sensed information at it to be as fresh (of small age) as possible, given network constraints. We propose the Age Control Protocol (ACP), which, in a network-transparent manner, enables a source to keep the age at the monitor small. We evaluate it using simulations and real-world experiments. ACM Reference Format: Tanya Shreedhar, Sanjit K. Kaul, and Roy D. Yates. 2018. Poster: ACP: Age Control Protocol for Minimizing, Age of Information over the Internet. In The 24th Annual International Conference on Mobile Computing and Networking (MobiCom ’18), October 29-November 2, 2018, New Delhi, India. ACM, New York, NY, USA, 3 pages. https: //doi.org/10.1145/3241539.3267740

摘要：实时监控的特点是：源通过网络重复向监视器发送更新，监视器希望在给定的网络约束下检测到信息尽可能的新。我们提出了ACP，它能够使源保持信息在监视器的年龄尽可能短。我们用模拟实验和真实世界的实验来评估它。

1 INTRODUCTION Real-time monitoring applications span across areas like health care and natural environment monitoring. They are distinct from that of file transfer and real-time voice/video. Such applications have a sensor/Internet-of-Things (IoT) device (source) that sends sensed information (update) to a monitor. The monitor desires to have the most currently sensed information at any given time. However, the same may not be feasible given constraints imposed by the communications network. To elucidate, let’s say the source sends fresh updates at a rapid rate. However, the faster the rate of updates, the larger the throughput, and the larger is the average delay (Figure 1) with which an update is received at the monitor. This is explained by the larger resulting congestion in the network. As a result, the freshest update at the monitor at any given time will be stale, that is of large age. Figure 1 broadly captures the behavior of the metrics of throughput, delay, and age. The throughput increases linearly in the rate of updates. This leads to an increase in average packet delay. Large packet delays coincide with large average age. Large age is also seen for small throughputs (correspondingly, small rate of updates). At a low rate the monitor receives updates infrequently, and this increases the average age (staleness) of its most fresh update. Finally, observe that there exists a rate (and corresponding throughput) at which age is minimized.

简介：实时监控应用覆盖医疗保健和自然环境监控等领域。它们与文件传输和实时语音/视频应用不同，它们有一个传感器（source）不断发送感测的信息*（更新）给监视器。监视器想要在任何给定的时间都能收到最新感测到消息。然而，考虑到通信网络施加的约束，同样的方式是不可行的。为了阐明，让我们说消息来源以快速的速度发送新的更新。然而，发送更新的速度越快，吞吐量就越大，以及监视器收到更新的平均延迟就越大。这是由于网络拥塞会更严重。因此，在任何给定时间监视器上最新的更新将是陈旧的，即年龄较大。图一大体上捕获了吞吐量、延迟和老化三个指标的行为。吞吐量随更新速度线性增加。吞吐量增加导致平均分组延迟增加。大的分组延迟与大平均年龄一致。吞吐量较小的时候（相应的，更新速率也小）年龄也比较大。监视器以较低的速率不频繁的接受更新，这个增加了最新更新的平均年龄（陈旧性）。最后观察到存在一个使年龄最小的更新速率（以及相应的吞吐量）。

We propose the Age Control Protocol that resides in the transport layer and operates only on the end hosts. Figure 2 shows an end-to-end connection between two hosts, the IoT device, and the server, over the Internet. Sources at a device open ACP connections to their monitors. ACP belongs to the transport layer of the TCP/IP networking stack. Akin to the real-time transport protocol (RTP) [3] that supports voice/video applications, ACP also uses UDP for sending updates generated by the sources. Thus ACP doesn’t guarantee delivery of a source update. Inline with the goal of ACP to optimize freshness, sending a new update is better than retransmitting an older one.

我们提出了ACP协议，该协议存在与传输层以及只运行在终端主机上。图2展示了两个主机设备（物联网设备和服务器）之间通过网络的端到端连接。设备上的源打开与其监视器的ACP连接。ACP属于TCP/IP网络堆栈的传输层。与支持语音和视频应用的实时传输协议（RTP）类似，ACP同样用UDP来发送源产生的更新。因此ACP不保证提供源更新。与ACP优化更新新鲜程度的目标一致，发送一个新的更新比重传一个旧的更好。

ACP suggests to a source the rate at which it must send fresh updates such that the average age at the monitor is minimized. ACP adapts its suggestion to the perceived congestion in the Internet. By suggesting this desired rate to a source, ACP also limits congestion that may be introduced by an unnecessarily fast rate of updating by sources.

ACP向源建议一个更新发送的速率，该速率使更新在监视器的平均年龄最小。ACP根据感知到的网络拥塞程度来调整它的建议。通过建议想要的速率给源，ACP同样可以限制源发送更新的速率过快引入的拥塞。

Let u(t) be the time at which the freshest update at the monitor at time t was sensed at the source. The age ∆(t) at the monitor is the age t −u(t) of its freshest update at time t. An example sample function of the age process is shown in Figure 3. Age at the monitor increases linearly in the absence of updates. On reception of an update that is fresher than updates received in the past, the age at the monitor is reset to the age of the update (di − ai for update i in Figure 3), which is simply the time elapsed between its generation at the source and its reception at the monitor. We want to choose the rate λ (updates/second) that minimizes the expected value limt→∞ E[∆(t)] of age at the monitor, where the expectation is over any randomness in the network. The expectation is not usually known and must be estimated using measurements. Also, we would like to dynamically adapt this λ to nonstationarities in the network

翻译：假设u(t)是最新更新到达监视器的时间，而t是源发送该更新的时间。监视器显示显示的时间∆(t)最新更新的年龄 t −u(t)。一个实例样本函数如图三所示。监视器上年龄在没有更新的情况下线性增加。一收到比过去更新更新的更新，监视器上的年龄重置为该更新的年龄（图三所示更新i的di-ai），这只是它在源上生成和在监视器上接收之间经过的时间。我们想要选择速率λ (updates/second),它可以使limt→∞ E[∆(t)]最小，其中期望值超过网络中的任何随机性。通常这个期望不被人所知，但是必须用测量手段估计。同样，我们想要使这个λ动态适应网络的不稳定性。

2 WORKINGS OF THE ACP An update sent by the source ACP contains a timestamp ai that is the time update i was sent (see Figure 3). A more recent timestamp indicates a fresher update. The monitor ACP sends an ACK packet in response to every received update that is fresher than the previously received updates. An ACK for update i contains ai . ACK(s) that are received in a delayed manner that is after an ACK of a fresher update has been received, are discarded. ACP doesn’t assume any time synchronization between the source and the monitor.

翻译：ACP的工作

源ACP发送的更新包含一个时间戳ai，ai是更新i被发送的时间（图三所示）。一个更近的时间戳意味着一个更新的更新。监视ACP发送一个ACK包，响应每一个收到的比上次更新更新的更新。一个更新i的ACK报文包含ai。延迟收到的ACK被抛弃。ACP不承担任何源和监视器之间的时间同步。

令Δk为监视器上的更新年龄（tk-1，tk）的时间平均值的源ACP处的估计。要计算此平均值，源ACP必须在监视器上的间隔内构建其年龄样本函数的估计值（参见图3）。它知道源发送某个数据包的时间i。但是，它需要监视器接收到的时间，它接收到收到数据包i的ACK时的时间。在接收到ACK时，它将其年龄估计重置为i的RTT。
设Bk为（tk-1，tk）计算的积压的时间平均值。当源发送新更新时，积压增加1。当接收到对应于更新i的ACK时，更新i并且从当前的积压计数中移除任何比i更早的未确认的更新。

来源的ACP在tk处选择一个行动，该行动针对相对于（tk-1，tk）的（tk，tk + 1）平均积压的变化b * k + 1。这些动作可大致分为（a）加性增加（INC），加性减少（DEC）和乘法减少（MDEC）。这里MDEC对应于一组动作MDEC（γ），其中γ= 1,2，.... 。 ..我们有κ> 0是步长参数，我们根据经验选择。 ACP试图通过适当设置λk来实现。可以证明（我们跳过细节），λk=图4b总结了ACP如何选择其动作uk作为bk和δk的函数。

第一个条件指示更新速率使得更新分组经历大于最佳延迟。 ACP尝试乘法减少积压以减少拥塞延迟，并在此过程中迅速减少年龄。这种情况的每次连续发生（通过每次增加γ跟踪）都试图更积极地减少积压，这是2的更大功率。
第二个条件bk <0，δk<0捕获年龄和积压的减少。 ACP贪婪地旨在进一步减少积压，希望年龄也会减少。如果前一个动作这样做，它会尝试乘法减少。否则，它会尝试减少添加剂
在（a）bk <0，δk> 0或（b）bk> 0，δk<0的情况下，源ACP的目标是下一个控制时期的平均积压增加。第一个条件暗示更新率太低导致年龄增加。因此，ACP增加了积压。在第二个条件发生时，ACP贪婪地增加了积压。
当条件bk <0，δk> 0时，ACP检查先前的动作是否试图减少积压。如果是，并且如果积压的实际变化远小于期望值，则ACP会成倍地减少积压。这有助于对抗年龄增长实际上是因为拥挤程度增加的情况。