RDS Rightsizing by Cloudyn

April 3 2012 | by Boris Goldberg AWS, Cloud Cost Management, Cloud Optimization, Cloud Provisioning

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- Filament.io 0 Flares ×

As discussed in last week’s post, there are cloud cost factors in the services approach (IaaS vs. Pass) that have to be taken into consideration when deploying in the cloud. The PaaS model can present particular challenges in flexibility of pricing, but can also offer advantages to a business looking for a fully integrated software stack.

Cloudyn has spoken extensively about how AWS EC2 (Amazon’s IaaS approach) pricing can by managed with rightsizing and price optimization in the cloud. AWS EC2 offering has about 12 distinct sizes and 3 pricing models: On Demand, Reservations, and Spot. The reservation instances then in turn have several flavors, which can be grouped by two additional dimensions:  term (a 1 or 3 year contract) and Usage type (Heavy, Medium and Light).

Today, we’d like to walk you through the new AWS RDS (Amazon’s PaaS approach) capabilities that Cloudyn is rolling out to its customers this week. For AWS RDS the offerings are in much less distinct sizes and models. Spot instances are not available for RDS, but the On Demand and Reservations instances pricing models are quite similar. From a sizing standpoint, RDS has only six different distinct sizes, focused primarily on memory and CPU configurations.

As a relational database as a service, the process for optimizing costs for many RDS solutions has multiple components to be considered, such as physical and logical structure, write policy, buffer pool size, etc. The overall performance levels can be impacted with seemingly minor changes to the multiple parameters and thresholds, so all have to constantly be monitored.

Cloudyn has developed a series of rules for determining rates of utilization and parameters of when the downsizing to the next level would be optimal for a business.

In order to decide whether RDS is significantly and consistently underutilized we are examining 4 basic metrics: CPU Utilization, Freeable Memory , Read Throughput, and Write Throughput. For each distinct DB Instance Class (except the smallest one) we have a separate set of threshold and statistical aggregation, arranged as heuristic rules for each recommendation level.

For the AWS RDS cloud usage, Cloudyn analyzes the information available via Amazon’s CloudWatch to compile usage patterns of machine utilization, maximum memory and input and output rates to identify cases of dramatic under-utilization of resources. Amazon’s CloudWatch stores the customer use patterns for only two weeks of time, but we’ve been able to offer users a longer-term view of the usage by archiving this information. This data is pulled directly from CloudWatch via a customer’s read-only cloud credentials and is then stored, aggregated and analyzed frequently for changing utilization patterns, a process that helps enable long-term resource planning.

Real-life customer story

Here is one example of how this is already working for one of our customers, who is a heavy cloud spender and is using multiple RDS instances. Within less than two weeks of monitoring of “Customer X” instances, Cloudyn’s solution detected a significantly underutilized instance and a potential for a 41% savings, should it be downscaled to a smaller size instance.

Let’s look into the details of this case study:

Customer X has instance of type db.m2.xlarge RDS. The instance is powerful, with 6.5 Elastic Compute Units and 17.1 GB RM. Upon examination we found that the actual usage was much lower than the instance capacity.

For Customer X, usage levels were at:

  • Average CPU utilization of only 0.76%
  • Max CPU utilization at only 12.44% (in peaks)
  • Min Available RAM of the instance was 15GB (i.e. never used more than 2GB)
  • Average number of bytes read from disk per second was  13.25
  • Average number of bytes written to disk per second was  38,240.9

What this means simply is that this instance could be SAFELY scaled down to a db.m1.large with 4 ECUs and 7.5 GB RAM instance, without any impact on performance and while saving over 40% of its cloud costs.

Average customer savings we’ve seen with RDS is potentially in the range of 35-50% annually of current costs.

Cloudyn users are welcome to visit their dashboard and find relevant updated savings recommendations, and other cloud users are welcome to check out how much they can save by Starting our Free Trial.

Connect with us
Sign up for our newsletter
  • альтернативный текст
  • альтернативный текст
  • альтернативный текст
  • альтернативный текст

Login

SSO Login

Forgot Password?

No account yet? Register.

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- Filament.io 0 Flares ×