Enterprise Cloud Computing Requires Service-Level Discipline
Real Time Infrastructure Enables Clouds through Quality of Experience (QofE) Management
By: Tony Bishop
Dec. 11, 2009 01:00 PM
Adaptivity at Cloud Expo
Yet, the industry standard model and best practices for establishing, managing, and measuring contracts via traditional Service Level Agreements (SLAs) has not been successful.
In practice most agreements are based on simple Key Performance Indicators (KPIs) that are grounded in arcane technical concepts and not in the context of the customer consuming services. The result is artificial measurements that lead to IT having a ‘green’ dashboard, but lost business opportunity and irate customers.
The source of this disparity is the traditional agreement which is limited to guarantees of uptime, quantities of dedicated resources, and/or KPIs of utilization. This supply-driven view of the platform does not map back to business events and therefore has limited applicability. Furthermore, this traditional agreement model entirely breaks down when considering the infrastructure in support of a Real Time Enterprise (RTE) operating in the cloud.
To overcome these limitations, a simple truth is offered: the best measure of business performance in the context of a technical system is the balance between the experience of its users, its cost, and its efficiency: its Quality of Experience (QoE).
A successful agreement will act as a lens on the desired QoE as defined by the business and define the policy and controls required to properly measure and manage services provided.
Forging Successful Agreements Based on Quality of Experience
Furthermore, this framework not only provides an introduction to ELAs, but also adapts the traditional SLA and OLA in order to measure and manage the qualities most important to supporting business demand within a real-time and dynamic operating environment.
Using the new lens of QoE, there are generally three types of agreements that can be forged between an IT consumer and provider.
* Service Level Agreement (SLA)—An agreement, not to be confused with the traditional SLA based on KPIs, that specifies the worst-case limit to the usage characteristics (demand) the consumer will generate and the lower limits of infrastructure characteristics (supply) the provider will provide in context of QoE. The document spells out the consequences of a breach by either party, which can generally be summed up as follows. If the provider is unable meet the service level, then the document spells out the consequences (frequently financial); however, if the upper consumption limits are exceeded, the provider commitment converts to a best effort to meet the demand. This type of agreement is a two-way commitment between the IT entity and its consumer. When entered into with an external entity it may be considered a legally binding contract.
This interpretation of a SLA is based on a different foundation than the traditional SLA. The agreement is two-way, which means it’s possible for either party to breach the SLA. Secondly, the basis for measuring SLA compliance is based on measurable characteristics the business deems important (e.g. response time, transaction time, quantities of users, etc.).
* Operational Level Agreement (OLA)—An OLA is very similar to a SLA, but between internal support groups it is used to solicit guarantees of performance from IT infrastructure components (web service, grid, middleware, storage, network, etc.) that the application or service is dependent on and to formalize relationships and coordinate support efforts. These components may also aggregate multiple resources, each of which can have its own OLA.
This type of agreement is also two-way, in a similar manner as the SLA. OLA compliance is also based on measurable characteristics that are in support of meeting the ultimate business needs (e.g., response time, transaction time, quantities of users, etc.).
* Execution Level Agreement (ELA)—With the advent of Real-Time Infrastructures (RTI), resources are no longer statically assigned to specific tasks. The ELA formalizes the agreement around how resources are allocated dynamically (high-priority systems get resources first, sometimes at the expense of lower-priority systems). The basic premise is that resources are not owned by the consumer, with the goal of right sizing the environment and that there are enough resources to do the job, while eliminating IT waste. In consideration for releasing the resources the consumer gets the added benefit of reduced cost and if their environment spikes unexpectedly the load will likely be manageable. ELAs are supplemental to a SLA or OLA, as the same two parties enter the agreement.
The concept of an ELA is a radical departure from traditional SLAs, which presume a static infrastructure. The flexibility of being able to change the allocation of resources based on business need, commonly referred to as a utility, can dramatically improve IT performance. To capitalize on the opportunity, this new governance enabled by the ELA is needed.
Implemented properly, SLAs, OLAs, and ELAs act as a link between business and IT that provides a means to measure the success (or quality) of IT services. They define the rules that dynamically and proactively apply the proper amount of resources as needed and when needed. At an elementary level, they define escalation rules in the event of service degradation or disruption. The result enables business success and delivers improved IT value to the company.
Creating an Agreement
Requirements need to be gathered at multiple levels of the system hierarchy: individual user experience (e.g., the home page will render in less than 500ms), sub-system interactions (e.g., data transfers will occur at a rate of 200 messages/second), and holistic (e.g., system activity may peak up to 50 active users). Each requirement must map to a metric that is measurable by IT but oriented around QoE. This can include KPI measurements so long as they map to user interactions or sub-system interactions. For example, measuring memory or CPU utilization may be useful for determining why a SLA is being missed, but are not likely part of the SLA guarantee. The SLA definition should be broad in scope and business goal driven, not limited to technology concerns.
Once the business driven aspects of the SLA are well understood, the IT organization should determine under what conditions the customer goals can be met, which become constraints within the SLA. If the consumer violates the constraints, then the SLA is invalidated. For example, consider the case where the business requirement states that the peak active users will be 50 and the system needs to have a response time of 500ms for the home page. Furthermore, IT stipulates there’s a constraint of 200 max active users to meet this goal. When the 201st user becomes active, the SLA guarantee no longer applies. This makes the SLA a two-way agreement between consumer and provider. While many constraints are soft limits, such as concurrent active users, others are hard limits, such as license counts. It simply isn’t reasonable to have the expectation that no constraints exist.
In our example, if there is a need to support 200+ users, then that should be a requirement and the system needs to be architected and engineered to comfortably support that load. RTI clouds can help deploy additional resources only to a point before the fundamental architecture of a system breaks down.
Quantifying Quality of Experience
Achieving QoE means balancing the user experience (e.g., response time, time to market, availability) with cost and efficiency. QoE cannot be achieved by continually spending money on infrastructure as problems arise. Understanding what the business needs and being able to measure those attributes are critical to achieving QoE. Without measurement there is no opportunity to control the environment to meet demand in real-time.
There are many categories of qualities that can be considered, and not all categories will be appropriate for all lines of business. Within these categories there are many quality attributes, and not all of them are applicable for every service or application. Here are some of the key categories for qualities; the nature of your environment may have specific needs so this list should not be considered comprehensive.
Scope of Agreement
Enacting the Agreement
Furthermore, the priorities of any OLA need to be evaluated against the overall SLA. If a system has high priority, but a sub-system has low priority, this can cause unnecessary conflict during recovery (and may be at the root cause of the failure). Any relationships between SLAs, OLAs, and ELAs must be clearly documented and preferably associated with the system in the Enterprise Configuration Management DataBase (CMDB).
Once in production, regular real-time tracking and reporting of SLA metrics and their evaluation against the configured limits becomes critical. This information can be used to manage routine capacity planning and give advanced warning of a system that may be approaching its architectural limits. Often the trending of this information will provide more insight than the raw numbers themselves.
In addition, the system will require automated monitoring. Monitoring is performed on various qualities of the services that comprise the SLA, including but not limited to end-to-end transaction monitoring, application monitoring, security service monitoring, IP connectivity monitoring, etc. These collected statistics, in relationship to the SLA qualities, should be presented in real-time back to the business through a dashboard. Note that the characteristics being measured are business driven (and not technology driven). So, the dashboard will likely provide significant new insights into business operations beyond the monitoring activities.
The monitoring and dashboard of the environment can provide real-time insight into this usage, providing a new level of transparency to the business, such that the business understands the quantity of the resources their actions are consuming. This provides a healthy environment within which to relate business actions with infrastructure expenses. This information can be used to help with justifying infrastructure expenses when making budgeting decisions around systems and services, which has traditionally been poorly understood.
The introduction of a new agreement framework based on Quality of Experience fundamentally changes how agreement is reached and defined. The conversation becomes oriented around business goals, is defined by qualities desired, and is agreed to as equal partners. The buy-in of both parties around a common language sets a reasonable foundation for moving forward.
The introduction of shared resource environments has impacts on datacenter operations. Prioritization of systems and services is required across a business line to ensure that dynamic resource allocation is policy driven based on business need, such that unexpected events are handled properly. The new ELA construct is a key enabler to managing the environment in a cost effective predicable manner.
Reader Feedback: Page 1 of 1
Latest Cloud Developer Stories
Subscribe to the World's Most Powerful Newsletters
Subscribe to Our Rss Feeds & Get Your SYS-CON News Live!
SYS-CON Featured Whitepapers
Most Read This Week