logo

In my last couple of posts, I discussed the journey that we are on and how applications will be purchased and deployed going forward. In this post, I'll discuss how I see systems being provisioned to support these applications. In order to support an environment where applications are purchased as a service, with performance and availability attributes, the underlying platforms must be very flexible and have advanced capabilities such as Quality of Service and High Availability. First, on the flexible part... it's no longer good enough to buy hardware based on the application to be developed. This is the old model and frankly it got us where we are today - with lots of underutilized, inflexible silos of hardware. The new model mandates hardware as building blocks. This doesn't imply that one size fits all - in fact, nothing can be further from the truth. What it does imply is that the hardware can be identified and configured automatically and remotely. It implies that hardware can be re-purposed for many uses. And finally, it implies that hardware, when identified, will be able to communicate its attributes so that the appropriate applications can be deployed to the best-fit hardware. These hardware attributes can be as simple as # of sockets and memory size, or can be as complicated as - does the hardware support redundant power and redundant devices, does the hardware support RDMA NICs or QoS capabilities, etc. This is obviously a big change to how applications and their host servers are deployed today. However, the underpinnings are there to support such a deployment method. 10G Ethernet is starting to gain traction in the data center and this can be used as a converged fabric (see Sadry's post on this). Hypervisors are starting to flow QoS networking information up to the guest machines running on them. So, the underlying platforms have the basic capabilities, but those alone can't make the correlation between the applications and the servers. This requires some advanced policy and/or automation software, which can understand the business process, the application requirements, and match those to the best-fit underlying hardware platform. So, there you have it. That's how I see applications being purchased, deployed, and run on a dynamic data center environment. Sounds a bit like the cloud vision, but it is a bit more complex and mission critical than your average cloud, which tend to like homogeneous hardware and an all-virtual environment. Thoughts? Leave a comment or send me an e-mail.