One of the promises of the cloud is better efficiency and utilization of infrastructure. Towards this end, Azure enhanced an Azure Advisor recommendation called “Shutdown/Resize your virtual machines”. It provides customers with cost savings opportunities by targeting VMs that are not utilized efficiently. Beforethe Advisor recommendation for right-sizing VMs had been created, customers needed to be constantly on the lookout for threshold breaches that involved balancing performance and costs. Now customers can get intelligent, proactive notifications if capacity rates aren’t high enough, making the recommendation more actionable and provide higher savings than ever.
Along with this enhancement, we’ve made a number of changes that increase the saving opportunity for customers as well as quality and actionability of recommendations.
Some of our bigger improvements are:
Cross SKU family series resize recommendations: up until this release, the resize recommendations that we had provided was mainly within the same SKU family. I.e. if you were using a D3v2 inefficiently, we would recommend a D2v2 or a D1v2, i.e. a smaller SKU but within the same family. We noticed that opportunity for much larger savings by moving across SKU family such that you use a SKU that perfectly fits your workload. For e.g. a workload might be running on a D8_v3, which is a 8 core machine with 32 GB, mainly because of the need for 32GB of memory. If we identify that this workload is mainly memory heavy and doesn’t use the CPU as much, we can recommend that the workload be moved to a E4_v3, which also has 32GBs of memory but has only 4 cores. By moving your memory intensive workload to a memory optimized SKU family such as the E series, you will avoid paying for those extra 4 cores that your workload didn’t need.
Cross version resize recommendations: In addition to finding the best fit across SKU families, we also find the best fit across versions. In general, newer versions of SKU families are more optimized, provide more features and better performance/cost than older versions. If we find that your workload is running on an older version and can get cost benefits without impacting performance on a newer version, then we will make that recommendation for the VM. For example, we might generate a recommendation to resize a D3_v2 VM to a D3_v3 VM.
Quality improvements: We received feedback that some of our recommendations were not actionable due to certain criteria not being considered. As part of this release we’ve taken this feedback to heart and considered even more SKU characteristics such as accelerated networking support, premium storage support, availability in a region, inclusion in an availability set, etc. to ensure our recommendations are of the highest quality. We also completely revamped our entire recommendation engine to be based on state-of-the-art machine learning algorithms to increase the quality, robustness and applicability of our recommendations
With these changes you should see an increase in savings opportunities thanks to our cross series/version recommendations, with some of the recommendations removed due to the addition of new constraints. For more details go to the docs page.
We thank you for the feedback that you’ve shared that made these changes possible. Do continue to send in the feedback on these recommendations. You can send in/vote for your feature requests at the bottom of this link.