6 Comments

Summary:

When we consider cloud services — and the apparent lack of significant quality of service (QoS) control in them — we must also consider the relationship between QoS and capacity. Specifically, with sufficient capacity, do we even care about QoS controls? Is it a question of […]

structure_speaker_seriesWhen we consider cloud services — and the apparent lack of significant quality of service (QoS) control in them — we must also consider the relationship between QoS and capacity. Specifically, with sufficient capacity, do we even care about QoS controls?

Is it a question of resources?

For as long as we have been building computing systems, we’ve been see-sawing between the poles of plentiful and constrained resources. When resources are plentiful, we develop new abstractions to encapsulate design changes, increase reuse and improve user experience. These new approaches invariably consume more resources than the approaches they replace. And what good is capacity if you don’t use it? During these times, QoS controls seem like an engineering extravagance involving needless complexity. The consequence of ignoring QoS resource limitations is predictable — plentiful resources eventually become constrained resources and another round of capacity increases is desired.

During constrained resource times, we get more particular about how we want our systems to operate and attempt to introduce concepts, such as priority, scheduling, pre-emption and queuing, to achieve specific behaviors. Within a single application, one can usually design the appropriate behavior into the application itself if specific platform or framework controls are lacking. As we move to sharing resources within a single governance environment (such as the typical enterprise operating a virtualized infrastructure), this type of “application-level QoS” approach usually doesn’t work as well and platform-level controls are desired. Even when these are lacking, however, management oversight and shared goals motivate all involved to work together to design acceptable cooperative resource partitioning, often with good results.

Interestingly, we can’t generalize this QoS approach to cloud services. If we look at true multitenant clouds, assuring enterprises of the presence of resource controls in shared platforms is desirable whether resources are plentiful or not. Put another way, big clouds with a lot of capacity don’t automatically assure acceptable application performance. Enterprises want resource guarantees for some applications, the presence of which assures consistent performance independent of the activities of other tenants in the environment.

So yes, we do require QoS controls in our application infrastructures, regardless of the level of resource contention, when delivering services in multitenant environments. With QoS controls available, cloud providers can offer a range of services and price points that provide more choice to customers and back these services with service-level agreements (SLAs) that go beyond uptime and mean time to repair (MTTR) specifications. The result for enterprises is lower-cost IT infrastructure, applicable to a greater range of application types, obtained by combining shared platform economics with high levels of performance assurance.

The right building blocks?

For years, the IT industry has been optimizing its IT hardware designs for deployments aimed at the typical enterprise application, specifically applications with dedicated infrastructure. In these deployments, established boundaries exist around common functions, such as switches, routers, servers, and storage arrays and many suppliers have emerged to compete in these respective product component categories. We have developed integration standards to ensure that these products can work together. At this point, the components of IT infrastructure for dedicated application infrastructure are mature.

But are these the right building blocks systems for the next generation of IT applications, especially when there is a high probability that these applications will be hosted in virtualized, multitenant cloud environments? There is no question that the abstractions are useful and should remain, i.e., a “server,” “firewall” and “switch” are good encapsulations of recurring deployment functions, but in many cases QoS controls are insufficient or missing. In my third and final post of this series, I’ll discuss market forces and vendor activities combining to bring another level of control to the cloud and speculate on what this might mean for the next generation of IT components.

This is Part 2 of a 3-part series. Please also see Part 1, Cloud Computing: A System of Control. The third and final post will run tomorrow.

Bryan Doerr is chief technology officer for Savvis.

You’re subscribed! If you like, you can update your settings

  1. Cloud Computing: Building Blocks for the Enterprise – Gigaom.com « AddingInfo.com Saturday, May 30, 2009

    [...] plentiful, we develop new abstractions to encapsulate design changes, increase reuse and … Read Full Post: Cloud Computing: Building Blocks for the Enterprise – Gigaom.com Related Info:Cloud Computing: A System of Control – Gigaom.comCloud Computing: A System of Control [...]

  2. jasonspalace Saturday, May 30, 2009

    in addition to cloud QoS controls, i believe we can also introduce another element of capital efficiency by maturing our use of bandwidth during off peak times.

    this is a great series, i appreciate the knowledge looking at the future of enterprise cloud deployment.

    – jason nadaf

  3. Cloud Computing: Enter the “Stacker” Sunday, May 31, 2009

    [...] This is the third post in a 3-part series. Please also see Part 1, Cloud Computing: A System of Control, and Part 2, Cloud Computing: Building Blocks for the Enterprise. [...]

  4. Avoiding Latency in the Cloud Saturday, June 20, 2009

    [...] and such solutions can be easily integrated into cloud services. When adopting cloud services, optimizing wide area network bandwidth use should be at the top of anyone’s checklist. Cloud providers are uniquely poised to significantly [...]

  5. Douglas Wilson Sunday, June 21, 2009

    I know high-grade B.S. when I see it. This is corporate techno-speak B.S. It takes up a page instead of a short paragraph. Look, dude, this is about uploading and downloading a lot of stuff fast and reliably. When you figure out how to do it, or who is doing it in a first-rate way, get back to us. Otherwise save the electrons.

  6. You might be interested in our initiative to bring cost awareness and quality service right up into the application runtime stacks abstracting the concept of a resource to something more more that networking queues using a metering engine that is based on activity based costing where cost can be latency, liability, leasing….

    OpenCore 6.1: QoS for Apps and CARS (Cost Aware Runtime & Services)
    http://opencore.jinspired.com/?p=2257

    OpenCore 6.0: Software Service Resource Metering
    http://opencore.jinspired.com/?p=1799

    We are hoping for standardization in the Java runtime and then to other languages, runtimes and platforms.

Comments have been disabled for this post