在OpenStack上使用分散式路由器設計與實作具備高可用性的異質負載均衡器架構

在OpenStack中，為了避免網路節點成為OpenStack架構的瓶頸以及Single Point of Failure，OpenStack自身提出了分散式路由器的概念，藉由把虛擬路由器分散到各個運算節點上讓大部分虛擬機的網路不再需要經過網路節點，從而有效地減少了網路節點的負擔。但是在OpenStack目前的架構中，當我們想要使用負載均衡服務時，流量卻無法透過分散式路由這個功能直接從網際網路進入服務器所在的運算節點上，而是需要先進入LBaaS Namespace所在的網路節點或是Octavia amphorae虛擬機所在的運算節點，之後才能被分流到其他運算節點上的服務器，分散式路由的功能無法有效進行作用，網路節點(或某一個運算節點)依然會成為OpenStack環境的瓶頸。為了改善前述的問題，我們參考了Octavia的概念，然後修改成使用多個基於虛擬機而成的負載均衡器，並讓這些負載均衡器同時運行在所有運算節點上，形成active-active的高可用架構，流量因此會被分流到各個負載均衡器時並透過分散式路由的功能直接由各個運算節點進出網際網路，而不在需要集中至單一節點上。在這篇論文中我們將會根據上述的想法提出一個新的負載均衡架構並實作於OpenStack之上。

關鍵字

負載均衡器；高可用；分散式路由器

並列摘要

In order to prevent the network node from being the bottleneck and SPOF of the OpenStack architecture, OpenStack proposed the concept of distributed virtual router(DVR). By distributing the virtual router into compute nodes, most of the network traffic no longer needs to go through the network node and thus reduces the load of the network node. However, while using the load balancing service of OpenStack, traffic cannot go directly from the Internet to the computing node where the server is located. Instead, traffic needs go through LBaaS Namespace or Octavia amphorae VM before being distributed to servers by the load balancer. DVR cannot play a part and network node (or one of the compute nodes) will still be the bottleneck of OpenStack. To solve the issue, we refer to Octavia and propose an active-active high availability load balancing architecture. The new architecture will use multiple VM-based load balancers to run simultaneously on each compute node, so traffic visiting load balancers will be distributed to load balancers without centralizing through one node. In this paper we will implement the new load balancing architecture on OpenStack with DVR play in part.