Idle resources will always need to exist, especially in a cloud architecture

With IT focused on efficiency – for reduction in operating expenses and in the interests of creating a greener computing center – there’s a danger that we’ll attempt to achieve 100% efficiency. You know, the data center in which no compute resources are wasted; all are applied toward performing some task – whether administrative, revenue generating, development cycles, or business-related – and no machine is allowed to sit around idle.

Because, after all, idleness is the devil’s playground, isn’t it?

But before we decide to technically exorcise our data center of any idleness that might be laying around, consider the ramifications: if there’s no idle resources, how are you going to scale out?

Yeah, exactly. Without idle resources you can’t automatically scale out. In fact, a cloud computing model doesn’t work unless there are resources available (idle) that can be applied at any moment to an application in need of scaling out.

BALANCING IDLE RESOURCES WITH EFFICIENCY

What’s necessary in a cloud computing model is to be allocate idle resources in the most efficient way possible. That sounds paradoxical, but it actually isn’t. Let’s say that you’re running near 100% efficiency now but recognize that just in case you’d better have some idle resources available.

You can:

1. Take advantage of cloud bursting or cloud balancing, using external cloud providers as “overdraft protection.” 
   
   Cloud bursting enables an organization to scale out on-demand into the cloud, only when necessary. This keeps its own efficiency as high as possible but provides for availability just in case. Cloud balancing assumes the resources are always available and uses advanced global application delivery (load balancing) techniques to balance the distribution of requests between all available data centers, whether local or in the cloud.
   
2. Invest in additional servers to ensure you have extra resources available in case you need them.
   
   There’s something to be said for “being prepared” and the easiest way to do that is to slap a few more servers (or blades, whatever) into the data center just in case. This may decrease your “efficiency rating”, if you have one, but it’s easy enough to counter by asking whether that service level agreement should actually be considered important or can you use it to wrap up your leftover lunch tomorrow. Mentioning SLAs generally gets attention, especially if you imply you might not meet it unless something is done.
   
3. Invest in an application delivery solution – or take advantage of one you already have – to improve the efficiency of the servers and applications such that you free up idle resources on existing servers
   
   This is the one everyone overlooks, but it’s probably the best solution in terms of long-term ability to balance efficiency with scalability. That’s because it allows you to improve the efficiency of servers and applications, which means you can achieve higher VM densities or just have “idle” resources sitting around that can be used in the event you need to scale out. It’s the best long term solution because it’s likely if you’re scaling applications you already have a load balancing solution (and need it) and if you’re lucky it’ll be one of the ones that can be extended through modules without disruption to existing infrastructure/applications. 
   
4. Move every thing to a cloud provider
   
   I know, it’s probably not feasible, but for those organizations for which it is, this actually makes a lot of sense. Moving all applications and IT operations to a cloud provider would certainly ensure scalability and idle resources because well, that’s what they promise. The thing to consider here is cost and where the line between your budget and scale stops. Infinite scalability is a fallacy – you can only scale as far as your budget will allow.
   
5. Change the math
   
   Seriously, find a mathematician and have them derive a formula that incorporates the need for idle resources into the overall efficiency equation. If folks can prove 1=0 then you can find someone to come up with a formula that balances the two and still comes up with a 100% efficiency rating.
   
6. Ignore the problem and hope that “just in case” doesn’t happen. Make a note to sacrifice a live chicken next week as insurance.
   
   Do nothing is always a solution. It may not be a good one, or the right one, or the one that ensures you have a job next week, but it is a solution.

KEEP ON TOP OF THE GAME

If you aren’t running at or near 100% efficiency now, keep on top of that. Consider how you’re going to balance the need for idle resources as efficiency (i.e. utilization) continues to increase and do something about it sooner rather than later. Plan a strategy that will maintain performance and capacity while ensuring you have the resources available – regardless of where they might be located – to scale out.

If you haven’t started a virtualization and or cloud computing initiative then you should carefully consider what technology will not only enable you to move toward that goal, but that will assist in providing the best balance of idle resources with high efficiency as you move forward.

Related blogs & articles:

• How do you get the benefits of shared resources in a private cloud?
• What is server offload and why do I need it?
• Cloud Balancing, Cloud Bursting, and Intercloud
• To Boldly Go Where No Production Application Has Gone Before
• Building a Cloudbursting Capable Infrastructure
• Scalability Only One Half the Reliability Equation

Read the original blog entry...