The Desired State Solving the Data Center s N-Dimensional Challenge
Executive Summary To solve this fundamental problem in the softwaredefined age how to assure application performance while utilizing the environment as efficiently as possible requires a software-driven approach. The software-defined revolution began with virtualization. Virtualization abstracted physical infrastructure, reducing silos and increasing efficiency. The decoupling of application workloads from the physical infrastructure they run on has enabled workload mobility across hosts, clusters, data centers, and clouds (e.g. vmotion, Live Migration). Previously manual tasks can now be completed in software. While the software-defined revolution simplified infrastructure management, it has also created a need for software-defined control. It continuously presents new opportunities to assure quality of service and maximize return on IT investments. In order to capitalize on these opportunities, we must be able to control the environment in a state in which application performance is assured. Controlling the environment has become much more challenging and complex complex beyond human scale. To solve this fundamental problem in the software-defined age how to assure application performance while utilizing the environment as efficiently as possible requires a softwaredriven approach. ii
What is the Desired State? The goal of IT is to assure application performance. But we operate in a world with constraints, where assuring application performance typically runs contrary to maximizing efficiency. The challenge for today s data center administrators is to simultaneously achieve both: keep the environment in a state in which application performance is assured, while maximizing efficiency this is the Desired State.
The Desired State What is the Desired State? The Desired State is consists of numerous tradeoffs in the data center, none in of which can be addressed in isolation. Tradeoffs: A dynamic equilibrium in which workload demand is best satisfied by infrastructure supply. between budget and cost between resiliency, performance and agility between application performance and infrastructure utilization between workload QoS and sweating assets between compute, storage and network bandwidth between compute, storage and endpoint latencies between infrastructure constraints and business constraints between compute and storage between CPU, memory, IO, network, ready Qs, latency, etc. between application priorities among business priorities These tradeoffs are between huge, conflicting forces that pull the data center in different, opposing directions in real time, all the time. When all tradeoffs are simultaneously satisfied, the environment is in a Desired State a dynamic equilibrium in which workload demand is best satisfied by infrastructure supply. 4
The Desired State Why It s Hard to Achieve The key to controlling the environment in this equilibrium is to be able to continuously and simultaneously analyze all the tradeoffs that define the Desired State and make decisions about the environment. The Desired State is continuously fluctuating. The key to controlling the environment in this equilibrium is to be able to continuously and simultaneously analyze all the tradeoffs that define the Desired State and make decisions about the environment. Accomplishing this is very challenging because virtual and cloud environments are: Dynamic: Every workload demand impacts the pool of available resources and thus every other workload demand. The interdependencies between entities and across the stack make the environment extremely dynamic every action, every resource allocation, every workload placement decision has a cascading effect. Constantly Changing: Workload demand constantly fluctuates because endusers and application workload demands are constantly changing. N-Dimensional: Every layer of the stack consists of different resources, which add multiple dimensions that must be considered among multiple tradeoffs. And they must be satisfied across multiple entities in the data center in order to achieve a Desired State. 5
The Data Center s N-Dimensional Challenge In the data center each resource (CPU, Memory, IOPS, IO, etc.) on an entity (host, VM, application, container, datastore, etc.) is a dimension. The environment can be in an infinite number of states, which are a function of those dimensions. In other words, the state of an environment is a state in an N-dimensional space.
The Desired State The Data Center s N-Dimensional Challenge The environment can be in an infinite number of states, which are a function of the dimensions. Looking at a single dimension, for example CPU on one host, there is a range of CPU utilization on the host in which VMs and applications sitting on that host will get the resources they need to perform, while maximizing utilization in the host. This acceptable range defines a subset of acceptable states. Each additional host defines such a subset. The acceptable states across all hosts is the intersection of all of these subsets. This intersection is a narrower range of acceptable states in the environment and is smaller than each of the subsets. See Figures 1 and 2. Figure 1 Delay vs. Utilization on a Single Host: There are infinite possible utilization states between 0 and 100. Each state results in a certain amount of delay as shown by the curve. Along that curve there is a range in which the level of utilization, and subsequently delay, is acceptable. In other words, the state of an environment is a state in an N-dimensional space. 7
Figure 2 Delay vs. Utilization on Two Hosts: Add another host with its CPU and the range of potential acceptable states for both is where they intersect we want to assure compute resources and maximize efficiency across all hosts. Graphically, the acceptable subset in each dimension is a sphere in N-dimensional space. The intersection of all acceptable states of CPU utilization across hosts is also a sphere in that space. Similarly, the acceptable RAM, IOPS, IO, etc. utilization across hosts are spheres in that space. The intersection of all of these spheres is a very small sphere. A state in this intersection is a state in which performance is assured while utilization is maximized across these resources and data center entities. See Figure 3. Figure 3 Good States Across Four Resources: The intersection of acceptable states for just four resources, RAM, CPU, IO, and IOPS. 8
Each variable is an additional sphere and the intersection of all spheres is the subset of the Desired States. Hence, the Desired States is an intersection of N spheres in N-dimensional space. See Figure 4. Figure 4 A Desired State: The intersection of good states for all resources (dimensions) across all data center entities. Below, a Desired State based on seven dimensions. As previously noted, there are, in reality, numerous tradeoffs and each additional dimension narrows the intersection of good states. 9
Solving the Problem Solving the problem of assuring application performance while maximizing utilization requires figuring out all possible Desired States the intersection of N-spheres in N-dimensional space and which actions will keep the environment in a Desired State (a state in this intersection). Given that virtual and cloud environments are dynamic, constantly changing, and growing, the solution must be scalable and real-time. The solution must be software-driven.
The Desired State Abstraction is Key Again A starting point for a unified The only way to solve these tradeoffs and scale to meet these challenges is with a unified software-driven control platform. Software was created to solve complex, software-driven control process-driven problems. In the case of the data center control, it must be able to platform is a proper data State, and keep it there. model. A starting point for this platform is a proper data model. A data model that holistically analyze the environment, make decisions that drive it to a Desired provides a common abstraction across the layers of the IT stack from the application all the way down to the fabric, hiding the messy details of the managed environment yet exposing enough to sufficiently control and maintain the environment in a healthy state. The common data model enables software to become an analytic engine that continuously determines the Desired State and drives all the actions to control the environment in that state. VMTurbo s platform abstracts the environment as a marketplace of buyers and sellers that trade compute resources as commodities they do so independently, across the environment and through every layer of the stack. VMTurbo uses this common data model to control the environment in the Desired State in which application performance is assured while the environment is utilized as efficiently as possible. 11
All entities in the environment are buyers and sellers: the Hosts, Data Stores, VMs, Applications, Containers, Data Centers, Virtual Data Centers, Zones, Regions, Filers, etc. buy and sell compute resources. For example, the hosts sell memory, CPU, IO, Network, etc. The Data Stores sell IOPS, Latency, etc. The VMs buy these commodities and sell vmemory, vcpu, vstorage, etc. Additionally, constraints, configurations, and business requirements are commodities traded in the market, as well. an Invisible Hand in the data center market, which drives the environment to an equilibrium where the workload demand is best satisfied by the available capacity at every layer of the supply chain i.e. every layer of the IT stack. A Desired State Can t Be Achieved with a Collection of Tools Don t mistake the automation of any single decision by a tool whether by DRS, SDRS, reclaiming, right-sizing, or any other mechanism in isolation as a solution. Any approach that looks at a specific resource or subset of resources in isolation cannot drive the environment to a Desired State. Again, the Desired State is a dynamic equilibrium in which application performance is assured, while maximizing efficiency across all resources and entities within the data center. Achieving a Desired State is Control (Not Fire-Fighting) VMTurbo s algorithms apply the principles of supply, demand, and price to the data center. The sellers continuously price their resources as a function of their utilization. The buyers, at the same time, continuously shop for the best deal they can get for the resources they consume. Just as with economic markets, enabling entities in the data center to work out workload placement, sizing, and configuration among themselves creates Driving virtual and cloud environments to a Desired State and keeping it there is control. Traditional approaches of monitoring and alerting neither control the environment nor assure application performance. They leave the human with the heavy lifting of troubleshooting and fire fighting. 12
Conclusion While the software-defined revolution simplified infrastructure management, it created the complex challenge of controlling virtualized environments. Solving for the Desired State in which application performance is assured while maximizing efficiency is complex beyond human scale. It requires a software-driven approach, which in turn, requires another layer of abstraction.
The Desired State Conclusion No matter the changes and fluctuations, VMTurbo keeps your environment in the Desired State. Using its market abstraction, VMTurbo s unified software-driven control platform drives a broad set of actions to control the environment: workload initiation and termination, workload placement and configuration (initial and continuous), compute and storage provisioning and de-provisioning, compute and storage configuration, etc. Whether you are trying to control the current workload, about to on-board new workload, or are planning how the environment will look in a day, week, month or year from now no matter the changes and fluctuations, VMTurbo keeps your environment in the Desired State. 14
About VMTurbo VMTurbo s Demand-Driven Control platform enables customers to manage cloud and enterprise virtualization environments to assure application performance while maximizing resource utilization. VMTurbo s patented technology continuously analyzes application demand and adjusts configuration, resource allocation and workload placement to meet service levels and business goals. With this unique understanding into the dynamic interaction of demand and supply, VMTurbo is the only technology capable of controlling and maintaining an environment in a healthy state. The VMTurbo platform first launched in August 2010 and now has more than 40,000 users, including many of the world s leading money center banks, financial institutions, social and e-commerce sites, carriers and service providers. Using VMTurbo, our customers, including JP Morgan Chase, Travelport, and Thomson Reuters, ensure that applications get the resources they need to operate reliably, while utilizing their most valuable infrastructure and human resources most efficiently. 2015 VMTurbo, Inc. All Rights Reserves. All trademark names are the property of their respective companies. 15