10

VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to
Page 2: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

© 2014 VMTurbo, Inc. All Rights Reserved. VMT-WP-Control0114.

THE VMTURBO CLOUD CONTROL PLANE

Software-Driven Control for the Software-Defined Data Center

EXECUTIVE SUMMARY

The Software-Defined Datacenter (SDDC) has the potential to extend the agility, operational and capital benefits that virtualization has brought about by extending its reach throughout the datacenter stack. The ability to configure the datacenter “on the fly” alone however, will not fully deliver on those benefits to enterprises and service providers.

In this paper, we outline the need for “Software-Driven Control” – the intelligence or “control plane” that can take advantage of these new software-defined capabilities, enabling enterprises and service providers to bridge the gap between software-defined flexibility and the true business potential of the Software-Defined Datacenter.

Page 3: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

THE VMTURBO CLOUD CONTROL PLANE

!

! 2!

THE PROMISE OF THE SOFTWARE-DEFINED DATA CENTER (SDDC)

Virtualization brought about radical change in the datacenter by decoupling the logical from the physical and introducing so-called software “control points”. A control point is a software-defined “lever” to the infrastructure that allows its change and manipulation. Examples of control points in the compute and storage stacks would be VMware vMotion and Storage vMotion, which enable Virtual Machines to be moved seamlessly across physical hosts and datastores respectively. These control points introduced unprecedented flexibility into the virtualization / compute layer, enabling the potential to adjust resource allocation and workload placement like never before.

Furthermore, the DNA behind virtualization continues to spread to the storage (Software-Defined Storage) and network domains (Software-Defined Networking) as evidenced by the recent $1.2B acquisition of Nicira by VMware, so that software-enabled control points now encompass all facets of the datacenter technology stack – hence the “Software-Defined Datacenter” (SDDC).

In most virtualized datacenter or cloud deployments, it is not unreasonable to have thousands or even tens of thousands of control points across the datacenter technology stack that may be programmed and manipulated. The question that remains however, is how will these control points be leveraged to deliver the agility, operational and capital efficiencies desired by the business? In other words, the flexibility offered by software-defined control points is necessary, but not alone sufficient to deliver on the promise of the Software-Defined Datacenter. How will these newfound control points be leveraged effectively to assure application performance while utilizing the infrastructure resources as efficiently as possible?

DRIVING DESIRED BEHAVIOR THROUGH ABSTRACTION, ANALYTICS AND AUTOMATION

Solving the right problem is key to providing the desired agility, operational and capital efficiencies in the software-defined datacenter, in order to drive the desired behavior in the environment to achieve business goals. This is an Intelligent Workload Management problem, and can be simply stated as:

Assuring applications meet desired service levels while utilizing the datacenter resources as efficiently as possible

It may sound simple but the reality is that this is an insanely complex problem to solve. Addressing one side of the statement is relatively straightforward – for example, we may overprovision the environment in an attempt to try and assure performance – but clearly, this will not efficiently leverage underlying datacenter resources. Conversely, one might

Page 4: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

SOFTWARE-DRIVEN CONTROL FOR THE SOFTWARE-DEFINED DATA CENTER

3!

try to run the environment “hot” or “on the edge” with highly-utilized resources, which can impact application performance creating incidents and problems.

Traditional monitoring tools are typically built “bottom up” and not designed to drive the desired behavior to solve this problem. They collect and present reams of data, leaving the burden on the operator to try and interpret this information – let alone solve the problem in large dynamic environments, across multiple interdependent layers of the datacenter stack.

Computer Scientists believe that the right level of abstraction is key to solving many IT problems. The right approach to solving the Intelligent Workload Management problem and achieving the desired behavior is top-down – one that understands the control points that can be leveraged to tune the environment and uses only the data needed to prescribe the necessary actions to maintain the system in its desired operating state.

Doing this correctly requires a layer of abstraction across the environment through which an analytic model can be run to determine the right actions to take based on business rules and system interdependencies. This solves for “data collection at scale” issues that arise in larger environments and ensures that any actions are prescribed with full understanding of the topological relationships in the infrastructure. By focusing specifically on prescriptive analytics, this type of solution approaches operations management with the goal of preventing performance problems based on service-level priorities and determining the specific actions to allocate resources appropriately.

Approaching the Intelligent Workload Management problem top down also means starting with the applications, their resource requirements and relative priorities, and ensuring that the (software-defined) datacenter is orchestrated appropriately to meet these requirements. The right abstraction of the Software-Defined Datacenter therefore, is one that enables understanding of the needs and requirements of the applications it serves.

Solving the Intelligent Workload Management problem must also encompass both business and technical policies and constraints, such as restrictions on where (and when) applications may be run, disaster recovery policies, and technical constraints such as storage, compute and network boundaries. It therefore requires an analytics layer that assures application service levels while efficiently utilizing datacenter resources, while respecting the necessary business and technical constraints at all times.

Page 5: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

THE VMTURBO CLOUD CONTROL PLANE

4!

ANALYTICS

When we consider the Intelligent Workload Management problem, it may be recast as balancing supply and demand for resources. For example, bottlenecks are formed when local workload demands exceed local supply of resource capacity. This suggests the use of efficient market techniques to redistribute the demand, or increase the supply. Indeed, a large body of research has established the value of such economic techniques for IT resource management.1

Accordingly, VMTurbo’s resource- and performance-management technologies are based on an economic model involving two sets of abstractions:

• Modeling the virtualized datacenter IT stack as a service supply chain, wherecomponents (e.g., VMs) consume services of other components (e.g., physicalhosts) and offer services to their consumers (e.g., guest OSes, applications). Thissupply chain spans from the applications through to the underlying compute,storage and network resources in the datacenter stack, and even horizontallyacross private and public clouds.

• Using pricing mechanisms to balance the supply and demand of services along thissupply chain, resource services are priced to reflect imbalances between supplyand demand, and drive resource allocation decisions. For example, a bottleneck,reflecting excess demand over supply, will raise prices of the respective resource.Applications competing over the resource will shift their workloads to alternateresources to lower their costs, resolving the bottleneck. Application budgets mayalso be set appropriately to embody relative application priorities (e.g. missioncritical vs. non-mission critical), and ensure that the datacenter stack is thereforeappropriately orchestrated to meet them.

1See e.g. R Moreno and A. B. Alonso-Conde, “Job Scheduling and Resource Management Techniques in Economic Grid

Environments” (2004) in Lecture Notes in Computer Science, and D F Ferguson et al., “Economic models for allocating resources in computer systems” (1996) in Market-Based Control: A Paradigm for Distributed Resource Allocation.

Page 6: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

SOFTWARE-DRIVEN CONTROL FOR THE SOFTWARE-DEFINED DATA CENTER!

5!

!

Figure 1 - The Service Supply Chain

The service supply chain together with the pricing mechanisms described above result in a set of decisions/actions that drive the virtualized datacenter into equilibrium where performance is assured while resources are used as efficiently as possible, respecting business and technical constraints.

ORCHESTRATION – CLOSING THE LOOP

As we look to driving the desired behavior into the virtualized datacenter, this final step is critically important to build upon the abstraction and analytics layers above. While there have been management systems that have attempted to apply increasingly sophisticated analytics, very few if any have closed the loop in order to control the environment and keep it in a state of equilibrium.

Figure 2 - Closing the loop

Page 7: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

THE VMTURBO CLOUD CONTROL PLANE

!

! 6!

The term automation is used extensively in the marketing lexicon of all IT operations management vendors; and with good reason—manual tasks are labor intensive and prone to error. When actions can be automated, they should be (IT process or run-book automation solutions do just that). These solutions automate many of the discrete tasks associated with running the virtual data center. However, they do not solve for the complex decision-making requirements that most IT operators face in maintaining the Software-Defined Data Center at the speed and scale it enables.

Automating these decisions requires a deeper level of understanding beyond just how to procedurally execute a set of tasks. To effectively ensure performance, the solution must be capable of determining what tasks to carry out. Once such actions have been identified, automation capabilities are becoming readily available across the Software-Defined Datacenter stack.

VMware revolutionized computing with their Distributed Resource Scheduler (DRS), by exploiting the fluidity of VM movement across hosts to try to actively balance Memory and CPU demands at the compute layer. DRS, however, remains limited in its scope, and does not consider other key resource demands beyond CPU and memory, or orchestrate across other (interconnected) layers of the IT stack, or across clouds – fundamentally, it does not solve the Intelligent Workload Management problem. Further, it is limited to VMware deployments in a time of dramatic growth in the deployment of heterogeneous hypervisor and cloud technologies.

Closing the loop with control systems is commonplace in a wide range of industry applications. Take, for example, modern fighter aircraft: the reality is that modern aircraft have thousands of control points – and are even designed to fly aerodynamically unstable to achieve greater agility. These aircraft cannot be flown without a control system in place. Similarly, the Software-Defined Datacenter will not deliver on its promise without such a control system in place.

PUTTING IT TOGETHER

Combining the abstraction, analytics and orchestration layers outlined above provides a foundation or “control plane” for maintaining the Software-Defined Datacenter in a state assuring application performance and service levels while utilizing datacenter resources as efficiently as possible.

!

Page 8: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

SOFTWARE-DRIVEN CONTROL FOR THE SOFTWARE-DEFINED DATA CENTER!

7!

Figure 3 – Software Driven Control

This control plane spans from Applications through to the hypervisor compute and related logical storage and network presentation. It also extends to the storage layer “beneath

the virtualization layer” – driving resource allocation decisions in the underlying physical

and logical storage constructs such as storage volumes, aggregates, controllers and filers.

Modern converged fabrics such as Cisco UCS combine sever and network capabilities

into a common chassis, the programmability of which enable “just-in-time” infrastructure provisioning to balance supply of compute and network resources with application

demand. Finally, the control plane extends to private cloud abstractions such as Virtual

Data Centers (VDCs) that present logical pools of resources to their tenants for enterprises and service providers offering IaaS services.

The control plane may even be extended to hybrid clouds leveraging public clouds such as Amazon Web Services (AWS) or Microsoft Azure, enabling operations teams to

understand at any point in time what applications and workloads should be run internally

vs. externally to maintain ongoing performance and resource efficiency, subject to business policies and constraints.

It is key to point out that as we saw in the service supply chain earlier, these layers of the datacenter stack are all interconnected and interdependent. Thus, resource allocation

decisions made in one layer will have “knock on” effects in other layers. The application

of an efficient market model described above with the action of buyers and sellers across the interconnected service supply chain, drives the environment into a global equilibrium

– enabling “cross domain control” of the Software-Defined Datacenter.

Page 9: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

THE VMTURBO CLOUD CONTROL PLANE

8!

CONCLUSION

This paper reviewed the tremendous potential of the Software-Defined Datacenter to increase agility and decrease operational and capital expenditures. We have concluded that the Software-Defined Datacenter needs a Software-Driven Control plane that provides the necessary intelligence, in order to take advantage of the newfound software flexibility to provide these benefits.

Using VMTurbo’s Software-Driven Control, enterprises and service providers can:

• Realize the operational, capital and agility benefits promised by the Software-Defined Datacenter

• Drive the datacenter into a desired state where application service levels are metwhile maximizing utilization by closing the loop between monitoring, analyzing andacting

• Take a top-down approach starting with application needs to orchestrate theunderlying compute, network and storage assets – within and across datacenter,private and public clouds

Page 10: VMT WP Control Plane for the Cloud - Updated€¦ · provisioning to balance supply of compute and network resources with application demand. Finally, the control plane extends to

����%R\OVWRQ�6W����WK�)ORRU��%RVWRQ�0$������

ABOUT VMTURBO

Founded in 2009, VMTurbo is a company founded on the belief that IT operations

management needs to be fundamentally changed to allow your organization to unlock the

full value of today’s virtualized infrastructure and cloud services. Our charter is to transform IT

operations in cloud and virtualized environments from a complex, labor intensive, and

volatile process to one that is simple, automated and predictable—delivering greater

control in maintaining a healthy state and consistent service delivery.

VMTurbo offers an innovative control system for virtualized data centers. By leveraging the

dynamic resource allocation abilities of virtualization and automating decisions for resource

allocation and workload placement in software, our solution ensures applications get the

resources required while maximizing utilization of IT assets. Over 9,000 enterprises worldwide

have selected VMTurbo, including British Telecom, Colgate, CSC and the London School of

Economics.

VMTurbo is headquartered in Massachusetts, with offices in New York, California, United Kingdom and Israel.

www.vmturbo.com