Most Read This Week
Why Intelligent VM Routing Is Critical to Your Private Cloud’s Success
Hosting decisions are far too important to be left to simplistic, best-efforts approaches
By: Andrew Hillier
Mar. 15, 2014 09:15 AM
Virtualized and private cloud infrastructures are all about sharing resources - compute, storage and network. Optimizing these environments comes down to the ability to properly balance capacity supply and application demand. In practical terms, this means allocating the right amount of resources and putting workloads in the right places. These decisions are critical to ensuring performance, compliance and cost control.
Yet most organizations are using antiquated methods such as home-grown spreadsheets and best guesses to determine which infrastructure to host workloads on and how much capacity to allocate. Not only do these approaches hinder operational agility, but as hosting decisions become more and more complex, they are downright dangerous. The typical strategy employed to stave off risk is to over-provision infrastructure, and the thinking behind this is that having an excess of capacity on hand will ensure that enough resource is available to avoid any performance problems. This is not only expensive, but it actually doesn't prevent key operational issues and many of the performance and compliance issues that are caused by incorrectly combining workloads.
In essence, this management challenge is the same one faced by hotel operators. Hoteliers need to constantly align guest demands with hotel resources and amenities. A hotel could not operate without a reservation system to manage resource availability and match that with guest needs, and yet this is exactly how companies manage their virtual and internal cloud environments. Imagine if a hotel didn't have the operational control provided by their reservation system, and was constantly forced to build more rooms than necessary in order to meet "potential" guest demands, rather than basing their decision on an actual profile of historical and predicted demand. Or if they put clients in rooms without enough beds or required amenities. This should start sounding familiar to anyone who has managed a production virtual environment.
Hotels have had the luxury of a long history to refine their operations, and by using reservations systems to properly place guests and manage current and future bookings, they have gained a complete picture of available resources at any point in time. In doing so, they have optimized their ability to plan for and leverage available capacity, achieving the right balance between supply and demand.
Why Workload Routing and Reservations are Important
1. Complexity of the Hosting Decision
2. Capacity Supply and Application Demand are Dynamic
Because of this, capacity must be viewed as a pipeline, with inbound demands, inbound supply side capacity, outbound demands and decommissioned capacity all being part of the natural flow of activity. Handling this flow is a key to achieving agility, which is a goal in the current breed of virtual and cloud hosting infrastructure. The ability to efficiently react to changing needs is critical, and the lack of agility in legacy environments is really a reflection of the fact that previous approaches did not operate as a pipeline. If it currently takes two to three months to get capacity, then it is a clear indication that there is no pipeline in place.
3. Meeting Your Customers Expectations
Sometimes the concept of a capacity reservation is equated with the draw-down on a pool of resources or a quota that has been assigned to a consumer or internal group. This is dangerous, as it simply ensures that a specific amount of resources will not be exceeded, and does not guarantee that actual resources will be available. This is analogous to getting a coupon from a store that says "limit 10 per customer" - it in no way guarantees that there will be any product left on the shelf. Organizations should beware of these types of reservations, as they can give a false sense of security.
Capacity reservations are extremely useful to those managing the infrastructure capacity. They provide an accurate model of the pipeline of demand, which allows for much more efficient, accurate and timely purchasing decisions. Simply put, less idle capacity needs to be left on the floor. It also allows infrastructure to be managed as a portfolio, and if a certain mix of resources is needed to satisfy the overall supply and demand balance (such as buying servers with more memory), then procurement can factor this in.
4. Even Self-Service Needs Reservations
The heart of the issue is a tendency for organizations to equate self-service with instant provisioning. Although instant provisioning is useful for dev/test, grid and other horizontal scaling scenarios, it is not the only approach. For example, an online hotel reservation site provides self-service access to hotel rooms, but these rooms are not often being booked for that night. For business trips, conferences and even vacations, you book ahead. The same process must be put into place for hosting workloads.
Rather than narrowly defining self-service as the immediate provisioning of capacity, it is better to focus on the intelligent provisioning of capacity, which may or may not be immediate. For enterprise workloads with proper planning cycles and typical lead times, reservations are far more important than instant provisioning. And deciding where the application should be hosted in the first place is a solution critical decision that is often overlooked. Unless an organization has only one hosting environment, the importance (and difficulty) of this should not be underestimated.
5. Demand Is Global
Operating in scale also allows certain assumptions to be challenged, such as the requirement for an application to be hosted at a specific geographical location. Geographical constraints should be fully understood and properly identified, and not simply assumed based on past activity or server-hugging paranoia. Some workloads do have specific jurisdictional constraints, compliance requirements or latency sensitivities, but many have a significant amount of leeway in this regard, and to constrain them unnecessarily ties up expensive data center resources.
Unfortunately, the manual processes and spreadsheet-based approaches in use in many organizations are simply not capable of operating at the necessary scale, and cannot properly model the true requirements and constraints of a workload. This not only means that decisions are made in an overly narrow context, but that the decisions that are made are likely wrong.
Moving Past Your "Gut"
Reader Feedback: Page 1 of 1
Subscribe to the World's Most Powerful Newsletters
Today's Top Reads