The good news about public cloud computing is that if you use it a little bit, or only on an intermittent basis, it is really cheap. The bad news is that if that casual use scales up to full time production use, public cloud computing can get really expensive in a hurry. This is especially a problem when people other than IT Operations sign up for the public cloud services – for example people who own and build applications in business units. Applications owners and developers do not have the cost minimization DNA that is very prevalent in IT Operations these days.
This has lead to two very interesting problems for people making use of public clouds. The first is that these people often do not have the skill set to “right-size” their images in the public cloud and tend to do what they always did on the physical side which is over-provision. The second problem is that these people often do not have the time to sit down and wade through all of the different price plans offered by the cloud vendor to figure out what combination of prepaid capacity and capacity bought on an as-needed basis makes the best combination of business and technical sense.
The Cloudyn Right-Sizing and Price Plan Selection Service
Cloudyn is a SaaS delivered service (it resides in a cloud itself), that collects data from the public API’s of the cloud provider. In its initial release, Cloudyn works with the CloudWatch API’s and is able to collect data about your usage of compute and database services from Amazon Web Services (AWS). What is interesting here is that Cloudyn is collecting more than commodity data about your CPU and memory usage. Cloudyn is figuring out how accurately (it turns out not so accurately for most people) you have sized your images (most people over-size them), and how your actual usage of the cloud services maps to the price plans you are using. Cloudyn then uses the optimization process depicted in the diagram below.
The Cloudyn Continuous Optimization Process
Some Initial Cloudyn Results
While in beta test, Cloudyn has been used by many customers to monitor over 50,000 Amazon images. During this trial phase Cloudyn has delivered an average of 40% savings to the initial set of customers. These savings have come in two forms. The first form is from right-sizing the images so that the customer stops over-provisioning. The second form is from matching the actual usage pattern of the cloud resources with the price plan that delivers the cloud services at the lowest possible cost. This process is depicted below.
Cloudyn Cost Savings
Issues (and Room for Improvement)
Cloudyn is the first service that allows people to optimize and control their costs for public cloud computing, and is therefore truly ground breaking in its own right. However as with the first release of anything there is always room for improvement.:
- Cloudyn does not install agents on your servers in the public cloud and by relying upon the public API’s of the cloud vendors uses a completely non-intrusive and no-overhead way to collect the data that it needs. However the data from the public cloud vendors is severely lacking in several respects.
- For example the Amazon CloudWatch data tells you how much of the resources you have been allocated you are using. That is great but that tells you nothing about the actual performance and transaction rate of your applications.
- What is really needed is a way for Cloudyn to know what response time your applications in the cloud must deliver at what transaction rate or load, and then optimize on that basis.
- Therefore the logical next step for Cloudyn is to partner with APM vendors who collect this data. Such a partnership will make both the APM cloud offerings from vendors like AppDynamics and New Relic more valuable, and make Cloudyn’s offering more valuable.
Cloudyn has delivered a breakthrough SaaS delivered service that does for the first time what no one else has done before. The new Cloudyn service actually tells you how to change your image provisioning and your price plans at Amazon so that you can achieve the results that you want at the lowest possible cost. This is the first time that someone has addressed the economics of public cloud computing in this manner. Ultimately this will lead to dramatically higher usage of public cloud services (as their price/performance can now be managed), and will put pressure on internal IT organizations to provide the same kind of data and management options to their internal constituents.