读了好些关于带宽分配的paper，多次碰到Hose Model这一概念。比如一下的论文：
[1] L. Popa, P. Yalagandula, S. Banerjee, and J. C. Mogul, “ElasticSwitch: Practical Work-Conserving Bandwidth Guarantees for Cloud Computing,” 2013.
[2] H. Ballani, K. Jang, T. Karagiannis, and C. Kim, “Chatty tenants and the cloud network sharing problem,” presented at the Proceedings of the 10th …, 2013.
[3] L. Popa, G. Kumar, M. Chowdhury, A. Krishnamurthy, S. Ratnasamy, and I. Stoica, “FairCloud: sharing the network in cloud computing,” presented at the SIGCOMM ‘12: Proceedings of the ACM SIGCOMM 2012 conference on Applications, technologies, architectures, and protocols for computer communication, New York, New York, USA, 2012, pp. 187–198.
[4] H. Ballani, P. Costa, T. Karagiannis, and A. Rowstron, “Towards predictable datacenter networks,” presented at the SIGCOMM ‘11: Proceedings of the ACM SIGCOMM 2011 conference, New York, New York, USA, 2011, p. 242.
[5] P. Soares, J. Santos, N. Tolia, D. Guedes, and Y. Turner, “Gatekeeper: Distributed Rate Control for Virtualized Datacenters,” 2010.
实际上这些论文里的Hose Model都是来源于以下SIGCOMM1999的一篇论文
N. G. Duffield, P. Goyal, A. Greenberg, and P. Mishra, “A flexible model for resource management in virtual private networks,” ACM SIGCOMM …, 1999.
Figure-1
A flexible model for resource management in virtual private networks 其实讲的是VPN下的资源分配模型。
最早VPN用是pipeline的方式，这样导致的问题是A，B，C，D四个客户相互连接的pipeline无法复用和灵活调整，客户需要精确地根据traffic matrix分配，因为每条pipeline都限制好了。比如A-D的线路，A-B的线路对于A来说，如果A-D的带宽暂时没用，A-B就没法用上A-D的空闲带宽。
The primary disadvantage of this approach is that it requires the customer to have precise knowledge of the traffic matrix between all the VPN sites. Resources made available to a customer-pipe cannot be allocated to other traffic.
如Figure-2：
Figure-2
因此有了Figure-1，通过软管这样的逻辑接口，可以让customer的出口和入口统一起来。控制软管的出口和进口，并提供性能保证。
In the hose model, a VPN customer specifies a set of endpoints to be connected with common endpoint-to-endpoint performance guarantees. The connectivity of each endpoint to the net- work is specified by a hose
Figure-3
因此到了目前的DCN里面，应用Hose Modle的场景就是这样的了：
Figure-4
每一条link上可以有多个flow，而这些flow在软管中的带宽是可以灵活调节的。另外这些link不一定是物理上的link，可能是逻辑上的link。
Hose Model于是体现在了DCN里面的链路里。
在SIGCOMM2013的这篇论文里便应用了Hose Model。简单的说其实就是把物理上的link聚合在一起看成可以灵活划分bandwidth的软管。实际上一个软管之下的物理链路是复杂的，但是通过软管能在逻辑的高度上灵活划分带宽。
这也就是为什么在SIGCOMM2011 的Towards Predictable Datacenter Networks 中说到：
The hose model is akin to the virtual cluster abstraction
Hose Model或者Virtual Cluster都是逻辑的抽象。
L. Popa, P. Yalagandula, S. Banerjee, and J. C. Mogul, “ElasticSwitch: Practical Work-Conserving Bandwidth Guarantees for Cloud Computing,” 2013.