These days, we hear a lot about cloud computing and how it's a massive shift from the traditional approach to a large-scale Internet-based service delivery environment. I'd like to shed some light on the technical aspects of this new model - hopefully providing some insight into the unique challenges we experience at Hostway.
From a basic infrastructure standpoint, not much has changed from the days of discrete hosting. The technological challenges in that arena have always been network reliability, power, cooling and scalability. The cloud model actually makes datacenter power delivery easier, as datacenter space experiences less churn while virtualization drives server consolidation. Server consolidation means less cooling is required and although density is higher, so is overall power efficiency.
The network, however, is radically different. In the datacenter, with multiple customers sharing the same hardware in most cases, a failed Ethernet port no longer means a single customer is offline. More than ever, absolute reliability is crucial. To this end, we now heavily rely on quality, rock-solid network devices and fast-acting, intelligent solutions to fault tolerance such as VSRP and Metro Ring Protocol in place of Spanning Tree.
Flexibility and scalability are just as important with the "have it your way, when you want it" nature of cloud hosting. Let's say a customer wants a number of instances, combined with a dedicated firewall and access to their portion of a shared iSCSI SAN. Not only must you have the flexibility to create this configuration, but the capacity to deliver it. Port density is very important in this environment.
Cloud hosting has also forced us to adjust our scope. No longer is a service restricted to a single datacenter or geography. The network must be able to reach further out to the consumer while providing overlay services, similar to CDN. In our case, this meant deploying our own multiservice transit network, capable of bringing our customers' data right to the edge while also bringing our datacenters closer together.
We chose to build our transit network on MPLS, with both IPv4 and IPv6. Traditionally a heavyweight technology, MPLS implementations are now mature, optimized and provide some very powerful advantages for large networks.
Our platform of choice is the NetIron XMR, which has density and configurability that allows us to deploy a much simpler POP. Wire-speed routing and switching on ports up to 10 gigabit Ethernet enable us to fend off DDoS attacks and handle unexpected surges in demand with grace. Just a few years ago, the same capability would have required a multitude of vendors and power-hungry devices.
It's been quite a journey the past few years and we're still evolving, but I'm excited to see where cloud computing takes us, both as an engineer and an Internet user.