Contributed article by Olivier Pomel, CEO and Co-Founder of Datadog.

Host-level Management becomes Irrelevant

As we look back on 2013, the shift to cloud computing dominated tech headlines. From our vantage point at Datadog, where a large number of our customers extensively use cloud-based infrastructure, we observed these movements to be quite tangible: A number of Global 2000 firms that we work with are abandoning their on-premise hardware to run applications on public clouds. Notably, this infrastructure automatically scales to match a customer's workload demand.

Auto-scaling cloud infrastructure provides engineering teams with more capabilities to grow computing resources. Engineers can now deploy "services" that are comprised of multiple hosts. These services automatically scale up or down at a moment's notice. Thus, specific hosts become trivial commodities. The performance of the aggregate is what matters. In 2014, this evolution will become complete and render host-level management irrelevant.

Public cloud infrastructure and automated configuration management as catalysts

Host-level management has been losing importance for some time. A decade ago, an IT department would agonize over which specific physical servers to buy for a project. Once installed, system administrators would monitor each server throughout its useful life. The widespread adoption of virtualization technology de-emphasized the physical server as hosts were decoupled from their hardware. Yet, virtual machines (VMs) were still managed as individual servers. VMs had to be deployed and overseen manually.

The ability for a public cloud to auto-deploy and parallelize an "infinite" amount of new servers made it feasible to level up to the "service". However, the catalyst for managing infrastructure at the service level has been the increasing adoption of automated configuration management systems such as Chef and Puppet. This software has allowed system administrators to liberally utilize auto-scaling while manageably overseeing countless hosts.

Infrastructure disposability and containerization accelerate this shift

As cloud auto-scaling gains increasing mainstream adoption, applications are being developed to take advantage of these features. Most notably, applications now treat infrastructure as disposable. Compute capacity can be built up or torn down at a moment's notice. Because few infrastructure components are ever permanent, it makes little sense to expend manual effort on any specific host. This new approach forces automated configuration management adoption. In turn, further use of automated configuration management produces a virtuous cycle, which increasingly enables auto-scaling.

Also, in 2013, new containerization projects such as Docker began to gain significant adoption. As this technology seeps into the mainstream, the host will become increasingly more marginalized. Containerization allows for applications and hosts to be recombined effortlessly, regardless of what on-premise data center or cloud the physical server is residing in. Thus, which host is being used by an application will be a negligible concern, as that host could change at any time.

The rest of the stack will have to adjust

A major consequence of host-level irrelevance is that the rest of the application stack will have to adjust. Systems that are downstream from the application will have to handle data from the infrastructure at the service level. In particular, monitoring systems will be forced to evolve. These systems have for decades required that new hosts be manually instrumented. Now, monitoring solutions will be left behind if they cannot auto-instrument new hosts which have been auto-deployed. Likewise, these systems' unit of measure will have to progress past individual servers so that performance can be assessed at the service level.

In conclusion, hosts in 2014 will become as irrelevant as processes in an application. Do you know how many processes are running your web app? No. It should be and will be the same with your hosts. The data center will scale-up one level.

##

About the Author