以太网速度发展现状

以太网联盟主席、戴尔首席技术官办公室以太网传播总负责人 John D’Ambrosia

2013年，以太网行业迎来了以太网40岁生日以及以太网标准诞生30年。以太网的速度进展一直是可预测的——以10倍的增量从最初的10Mb/s到100Mb/s，到1Gb/s，再到10 Gb/s，这几乎无可争议。不过，40GbE和100GbE的同时推出有效终结了这一传统。

40GbE和100 GbE的发展基于一个根本假设，即：计算与网络带宽的增长率存在相当大的差异，因此两种速度具有同时存在的必要性。在计算领域，带宽能力每24个月翻一倍，而网络应用程序则每18个月翻一倍。网络这种可预测的增长率后来同样被IEEE 802.3以太网带宽评估特别小组确认。据其预测，平均下来，到2015年，网络必须支持TB每秒的能力，到2020年，必须支持10TB每秒的能力。

下图显示了这两个应用领域可预测的带宽能力，这是由IEEE 802.3更高速以太网研究小组(后来发展成为开发40 GbE和100GbE的任务小组)在2007年底所做的预测。那么到了2013年，这一预测的准确程度如何呢?

图注：较早时候的以太网带宽发展预测

在现实中，支持和反对这一预测的说法都是可以成立的，但是否认同其精确度却取决于做出这一预测的思考角度。如果从电信运营商的角度来看，人们可能会认为这相当准确，因为100GbE在电信运营商领域产生了极大的影响，并且正在健康发展。但是在数据中心领域，却可以得出反对这一预测的结论，因为100GbE并没有在数据中心网络中产生同样大的影响——相反，40 GbE端口却在这里稳步健康发展。

这种新兴部署场景提出了一些需要考虑的问题：首先，为什么数据中心网络中部署的是40GbE，而不是100GbE?首先应该注意的是：数据中心之所以部署40GbE，是因为通过将4个10 Gb/秒运行的4个通道捆绑可以实现总共40 Gb/秒的聚合链路。这是一个重要的发现，因为40GbE端口与并行导体或光纤结合使用，就可以实现更高的四倍密度10GbE端口配置。

这就提出了一个有趣的问题——我们是否应将整个数据中心看作一个计算应用呢?如果我们考虑到40GbE部署的时机，并且看一看上图，就会发现，数据中心带宽需求似乎证明了那些最初认为40GbE适用于服务器的想法。如果我们接受这种推理，那么按照计算应用预测，意味着我们将在2017年~2018年看到数据中心部署100GbE。这一现象还提出了一个关于突破性功能的重大问题，这将对400G以太网的发展产生影响。

因此，当我们庆祝以太网40岁生日的活动进入尾声时，显然会得出结论，以太网将继续向前演进，尤其是在速度提升方面。这一判断提出了许多需要思考的问题，整个行业仍需要注重达成共识，以此推进以太网向前发展。

英文原文：

The State of Ethernet's Rate

by John D’Ambrosia, Chair of the Ethernet Alliance,Chief Ethernet Evangelist, Dell Networking CTO Office

In 2013 the Ethernet industry has been celebrating 40 years of Ethernet and 30 years of Ethernet standards. Ethernet's rate progression had been fairly predictable, 1_0x increments from its initial 10 Mb/s to 100 Mb/s to 1 Gb/s to 10 Gb/s with little to no controversy. The simultaneous introduction of 40 GbE and 100 GbE effectively ended this legacy.

The development of 40 GbE and 100 GbE was based on the fundamental assumption that the growth rates between computing and networking were sufficiently different to justify the two rates. For the computing space, bandwidth capabilities were doubling every 24 months, while network applications were doubling every 18 months. The predicted growth rate for networking was later re-confirmed by the IEEE 802.3 Ethernet Bandwidth Assessment Ad hoc, which fore_casted that on average, networks would need to support terabit per second capacities by 2015 and 10 terabit per second capacities by 2020.

Figure 1 shows the predicted bandwidth capacities of the two application spaces, which was made in late 2007 by the then IEEE 802.3 Higher Speed Ethernet Study Group (which would later go on to become the Task Force that developed 40 GbE and 100GbE). So in 2013 how accurate was this prediction?

Figure 1- Past Ethernet Projections

In reality, arguments for and against this prediction can be made, but that accuracy is truly dependent upon the perspective by which the prediction is being considered. If one looks at this prediction from the service provider perspective, one might argue that it was a fairly accurate, as 100GbE is having a big impact in the service provider space and enjoying healthy deployment. However, in the data center space, arguments against this prediction could be made, as 100GbE is not having the same impact in data center networks, which instead is seeing healthy deployment of 40 GbE ports.

This emerging deployment scenario raises a lot of issues for consideration. First, why is 40 GbE seeing deployment in data center networks, instead of 100GbE? The first thing that should be noted is that the 40GbE that is seeing deployment in the data center is based on 4 lanes running at 10 Gb/s to achieve an aggregate of 40 Gb/s. This is an important observation, because the 40GbE ports are being used with parallel conductors or fibers to achieve a higher quad density 10GbE port configuration.

This raises an interesting question – should we be looking at the entire data center as a compute application? If we consider the timing of 40GbE deployment, and we look at Figure 1, then the bandwidth requirements would seem to mirror those initially thought to be just for the servers. If we accept that reasoning, then following the predictions of the compute application fore_cast means we will see 100GbE deployed in the data centers in the 2017 to 2018 time frame. It also raises the question regarding the significance of breakout functionality, which would have implications for the development of 400 Gigabit Ethernet.

So as we come to the end of celebrating Ethernet's 40th Anniversary, it is clear that Ethernet will continue its evolution forward, particularly in respect to higher speeds. And while this perspective has raised many questions for consideration, the industry will need to focus on reaching consensus in order to drive this evolution forward.