Starting up a cloud computing deployment is a case of just choosing a cloud and switching it on, right? Kind of, yes. Cloud ‘instances’ can be ‘spun up’ inside 60 seconds -- but obviously it’s not quite that simple. Customers will need to choose between public, private (on-premises) or some hybrid combination of both.
CLUE: The answer to the title is the last line of this post.
Deeper still, cloud comes in many shapes, colors and flavors. Firms can opt for a cloud that has been ‘provisioned’ (or tuned, if you prefer) for specific performance characteristics such as:
- Storage optimized
- Compute processing optimized (with ‘burstable’ CPU even)
- Input/Output optimized
- Memory performance optimized
- Network connectivity optimized
Cloud providers today even offer a set of ‘general purpose’ instances. It’s not hard to draw an analogy between cloud computing and racing cars i.e. wet weather tires for specific conditions, off-road equipment for bumpy terrain… and so on.
It’s not called a ‘network’ anymore, anyway
So once your cloud Virtual Machine (VM) type is set, it’s fine to just slot it into the network, right? The answer is: kind of yes and no.
A rising degree of automation controls now manages what happens inside the cloud datacenter and it’s not called a network anymore anyway. Increasingly we now talk about the software-defined networking (SDN) environments that cloud computing instances are being engineered into. It’s still a network of course and there’s plenty of hardware involved, it’s just that intelligence sits in a more fluid software state.
But VM placements need a degree of their own analytics if they are to live happily ever after (or at least into the immediate future). This is the view of software-defined infrastructure control company Cirba -- as we have said before on Forbes, the firm’s analytics exist to automate what we call ‘routing decisions’ inside virtual machines. Cirba says it is currently seeing strong momentum for SDN adoption and the wider growth of the software-defined datacenter -- but then it would, wouldn’t it?
Can we fool network intelligence?
According to Cirba, “Most organizations are interested in SDN for the increased agility, provisioning speed and security it delivers. At the same time, however, it also puts greater pressure on the decision of where to place VMs within an environment. SDN abstracts the physical network layer and allows groups of VMs to be configured to ‘think’ they are on the same network segment, even if they are not. This means that a VM is no longer physically limited to residing on servers that are connected to a particular physical network. They can reside on any server within a broader mobility domain.”
The issue here is -- this control creates far more options as to which servers can host a particular workload. In turn then, this newfound freedom presents both opportunities and challenges in VM placement.
On one hand, SDN creates an increased potential for VM placements to violate business policies, compliance rules and perhaps software licensing boundaries. For example, if a VM running a database server is unconstrained in terms of where it is placed, most vendor licensing agreements will dictate that the full set of hosts must be licensed to run it. This necessitates a more rigorous approach to VM hosting and placement decisions that considers all the utilization requirements, alongside policies, technical requirements and software licensing considerations to determine VM placements.
“Cirba’s analytics optimizes VM placements by modeling all of the requirements and constraints affecting a set of workloads and uses these to determine optimal VM placements and resource allocations. In SDN infrastructure, this automated, policy-driven placement is critical to dealing with the complexity and lack of control that can arise,” said the firm, in a press statement.
Firms in the space generally include the bigger names such as VMware and BMC -- there are smaller names such as Evolven, CloudPhysics... but CloudPhysics provides placement specific to storage and Evolven is troubleshooting analytics for change and config management, not quite the same thing.
Intelligent VM placement will dovetail workloads in a way that is analogous to the game of Tetris.... [+]
It’s like a game of Tetris
In addition to avoiding risk, Cirba suggests this type of intelligent VM placement will dovetail workloads in a way that is analogous to the game of Tetris. It’s not just a case of turning the cloud on; it’s also a case of fitting it all together and clearing the road down which will we will travel.
It now becomes a question of optimization beyond the cloud cluster as we look to optimally place VMs on an enterprise-wide scale, effectively playing an even bigger game of Tetris -- and therefore making optimal use of the whole portfolio of infrastructure assets.
To answer our question when is a cloud not a cloud? When it is a non-optimized workload cloud that exists in a datacenter environment with zero traffic analytics management intelligence and no concept of resource allocation for the total software-defined network that it must exist within.
That’s when.
