The key proficiency for data centre managers is no longer hardware-centric - it’s applications and services based. Managing this is far more complex, as individual server hosts become more critically important - after all, they can be supporting multiple virtual machines, and moving images in advance of hardware changes becomes a vital job.
Virtualisation creates a significant disconnect between the methods of management of new virtual systems versus traditional physical infrastructure. This is no poor reflection on that infrastructure, of course - it’s often perfectly optimised for the pre-existing way of operating - but too much attention has been paid to virtualisation itself, and not the foundations that are being pressed into service to support it. The end result is that power failures are a major threat to virtualised computing.
Downtime is now acceptable
Let’s take a long, hard look at that statement. In the past, Five Nines reliability referred to hardware and operating environment. Nowadays, it is more allowable for physical kit to fail - so long as the application or service keeps on working, perhaps with slower response rates. The catch here is that, more often than not, virtual machines are treated like their physical counterparts, with the same types of management architectures. That means local management agents are applied to virtual machines in much the same way that they are installed on physical servers’ operating systems. This is simply not scalable.
Virtualisation brings to the fore the need for power management software. Back when infrastructure was designed from the get-go to be highly resilient for maximum uptime, this software was rarely needed, and therefore rarely specified by users. Nowadays, infrastructure is more loosely coupled to service - so the service needs to be smart to unplanned downtime and hardware failure.
Virtualisation creates complex structures
Moving from physical to virtual environments inevitably creates complexity, and as the load grows, more physical IT is added to support more virtual machines. Individual resources support multiple virtual machines. Managing how and where those machines are running, ensuring there are no duplicate versions or random instances, becomes vital. Inevitably, virtual machines are moved more frequently, too.
The problem is that traditional power management doesn’t work in this environment. Having one management console for the virtualisation and another for the power does not scale. Furthermore, deploying management agents to each virtual machine, keeping them functional and updated doesn’t make any sense when virtual machines can be spun up and down in mere minutes.
It makes far more sense to use the management console built into a the virtual environment - for example Microsoft System Center Virtual Machine Manager or VMware vCenter - and let it communicate with the power infrastructure via power management software rather than using separate unscalable agents on each virtual machine.
Where it made sense in the past to add more redundant UPS devices, generators and power feeds, it will eventually be more reasonable (and cost effective) to rely on one of the great things about virtualisation - that it’s not tied to a physical location. Instead of overprovisioning multiple redundant power supplies, it will eventually become possible to have more minimalist power backup in place, and be ready to move entire software-defined data centres to another set of hardware or even another facility.
Separating the physical and virtual
Allocating resources in the new virtualised environment looks clean and simple on screen, but it can be all too easy to forget that this virtual world is heavily reliant upon physical infrastructure designed for a different era. It is vital to understand how the physical and virtual coexist and interact, and how this affects the survivability of the services and applications provided.
One way of handling this is to avoid putting control agents designed for physical infrastructure server power management onto each virtual machine, but instead tie the power/UPS management software into the central management console, avoiding deploying and managing each unscaleable agent and the resultant IO bottlenecks. Aside from centralising management, it also allows the UPS management software to cleanly trigger the movement of virtual machines to another host if it spots potential downtime on the horizon.
In conclusion, it is vital to take potential problems for virtualised services to their logical and extreme conclusions. All too often, blind faith in central management servers of hypervisors drowns out any concerns about the physical plant that supports it. Similarly, applying traditional working practices to a virtualised environment misses out on the many benefits of having such an environment in the first place. Virtualisation is quite literally game-changing. But it needs to do so with an eye on the past, and an eye on the practical.
Naser Ali is Segment Marketing Manager at Eaton
Find your next job with techworld jobs