With due respect to the great bard – I couldn’t resist the play on words. I’ve been thinking about my comments last year on Cloud Computing and the reinvention of the bureau service 40 YEARS on: From MOON to CLOUDS: What’s HYPE, CREATION and REINVENTION? and a more recent article “IT chargeback: A fundamental part of cloud computing takes shape” by Laura Smith. I still believe that most of the “X as a Service” offerings are primarily complicated reinventions of the past – but with a more dynamic ‘contract’ for resource consumption and the legal framework required to protect data.
The issue of chargeback is obviously not new and Laura’s challenge/question is “what metrics can IT departments use to charge back business for cloud services?” Base the charging rate(s) using a typical utility model rather than the conventional model, and devise a metering system in the virtual world? To me this is where Analytics as a Service (AaaS) come into play – as the buying of disk space or running an application is not really difficult – same old model. But when you start to charge for the analytics performed where resource consumption is dependent on both CPU and storage you move into the realm of data warehousing/BI services.
Teradata has been delivering Private Cloud AaaS solutions for quite some time through the provision of Analytic Sandboxes or Data Labs, often on production systems. This is a space provided to end users who are able to load the data of their choice, combine with production data, and execute queries against it. Typical use of this service is for developing analytic models prior to promoting them into production.
So how does this additional [non-production and unscheduled] workload affect the overall performance of the production system? Laura points out that “Virtual servers are frequently oversubscribed, meaning that they probably would crash if all the clients on them required maximum compute power.” Two issues emerge here – guaranteed performance and what to charge if you don’t get the desired performance?
Most software services are offering list price per CPU licences not utility usage per hour pricing and most also allow you to bring your own licence (BYOL). The hardware service is usually charged per GB or per hour. For example, if you use a server for four hours but performance is horrible because of multi-client users, what price should you pay? Because with analytics – you want the query to not just finish, but to finish in a predictable and acceptable timeframe. You are consuming the same amount of CPU and disk – just the elapse time changes. Enter stage right the SLA! And stage left “T-Perf” resource consumption metric*. The SLA can apply to the rate per hour charged – similar to on and off peak electricity or mobile phone rate plans, the metric is the meter of usage.
A critical component to solving both the SLA and subsequent chargeback model is a rules based active workload management capability - something that Teradata customers have had the benefit of for years. The beauty of active workload management is that you can develop rules and assign different priorities to query type, user, data and, of course, time of day. The SLA drives the priority and access to resources. Once you can control resource access and capture user consumption all you then need to do is devise a rate plan for T-Perf per hour or second that takes into account the SLA guarantee ‘service’.
In fact – Teradata customers do have chargeback models based on T-Perf consumption based on statistics captured in logs. Billing/chargeback truly based on resource usage and over-layed with time-of-day or other characteristics such as priority or data ‘heat’ to simulate a utility billing/chargeback structure. Ideally it should include a base component for access. This model is then no different from electricity or water utilities or communication services (data/calls). You have a contracted connection fee (quarterly, monthly depending on the utility), and then a combination of a “per connect” fee (per call for mobiles) and then call fee based on time connected or kilo-watts of electricity consumed – both having different rates by time of day.
So as long as you have the consumption metric and the variable characteristics agreed to – voila! IT have a fair and equitable billing mechanism for “Analytics as a Service” – well at least in the Teradata world it can be (and is being) done today while others wait in hope.
"Hope is a waking dream."
- William Shakespeare, As You Like It, Epilogue
"
The post Chargeback for Analytics as a Service: or “AS you like IT” appeared first on International Blog.
