Top 11 Hard-Won Lessons We’ve Learned about AWS Auto Scaling
Auto scaling, as we know today, is one of the most powerful tools leveraging the elasticity feature of public cloud – Amazon Web Services (AWS). Its ability to improve the availability of an application or a service, while still keeping cloud infrastructure costs under check, has been applauded by many enterprises across verticals, be it fleet management services or NASA’s research base.
However, at times, AWS Auto Scaling can be a double-edged sword. For the reason that, it introduces higher level complexity in the technology architecture and daily operations management. Without the proper configuration and testing, it might do more harm than good. Even so, all these challenges can be nullified with few precautions. To this end, we’ve collated few lessons we learned over a period – to help you make the most of Auto Scaling capabilities on AWS.
Use Auto Scaling, whether your application is stateful or dynamic
There is a myth among many AWS users that AWS Auto Scaling is hard to use and not so useful with stateful applications. However, the fact is that it is not hard to use. You can get started in minutes, with few precautionary measures like using sticky sessions, keeping provisioning time to minimum, etc. Plus, AWS Auto Scaling helps monitor the instances and heals them if they become unhealthy.
Here’s how: Once the Auto Scaling is activated, it automatically creates an Auto Scaling Group, and provisions the instances accordingly behind the load balancer. This maintains the performance of the application. In addition, Auto Scaling’s Rebalance feature ensures that your capacity is automatically distributed among several availability zone to maximize the resilience of the application. So, whether your application is stateful or dynamic, AWS Auto Scaling helps maintain its performance irrespective of compute capacity demands.
Identify the metrics that impact the performance, during capacity planning
Identify the metrics for the constraining resources, like CPU utilization, memory utilization, of an application. By doing so, it will help track how the resources are impacting the performance of the application. And the result of this analysis will provide the threshold values that will help scale up and scale down the resources perfectly.
Configure AWS CloudWatch to track the identified metrics
The best way forward is to configure Auto Scaling with AWS CloudWatch so that you can fetch these metrics, as and when needed. Using CloudWatch, you can track the metrics in real-time. CloudWatch can be configured to launch the provisioning of an auto scaling group based on the state of a particular metric.
Understand functionality of Auto Scaling Groups while using Dynamic Auto Scaling
The resource configurations have to be specified in Auto Scaling groups feature provided by AWS. Auto scaling groups would also include rules defining circumstances under which the resources will be launched dynamically. AWS allows assigning the of autoscale groups to the Elastic Load Balancers (ELBs) so that the requests coming to the load balancers are routed to the newly deployed resources whenever they are commissioned.
Use Custom Metrics for Complex Auto Scaling Policies
A practical auto-scaling policy must include multiple metrics, instead of just one allowed by CloudWatch. The best approach to circumvent this restriction is to code a custom metric as a Boolean function using Python and the Boto framework. You can use application specific metric as well along with default metrics like memory utilization or CPU, network, etc.
Use Simple Queuing Services
As an alternative to writing complex code for the custom metric, you can also architect your applications to take requests from a Simple Queuing Service and enable CloudWatch to monitor the length of the queues to decide the scale of the computing environment based on the amount of items in the queue at a given time.
Create Custom Amazon Machine Images (AMIs)
To reduce the time taken to provision instances that contain many custom software (not included in the standard AMIs), you can create a custom AMI that contains the software components and libraries required to create the server instance.
Scaling up other AWS services other than EC2, like AWS DynamoDB
Along with AWS EC2, other resources such as AWS DynamoDB, can also be scaled up and scaled down using Auto Scaling. However, the implementation of the policies are different. Since storage is the second most important service other than computing service, efforts to optimize storage will yield good performance as well as cost benefits.
Predictive Analytics for Proactive Management
Setting up thresholds as described above is reactive. Hence, you can leverage time-series prediction analytics to identify patterns within the traffic logs and ensure that the resources are scaled up at pre-defined time, before events take place.
Custom define Auto Scaling policies & provision AZs capacity accordingly
Auto scaling policies must be defined based on the capacity needs as per Availability Zone (AZ) to save on cost spikes. Because pricing of the resources are based on different regions that encompass these AZs. This is critical especially for Auto Scaling groups configured to leverage multiple AZs along with a percent-based scaling policy.
Use Reactive Scaling policies on top of schedule scaling feature
By using Reactive Scaling policies on top of schedule scaling feature will give you the ability to really respond to the dynamic changing conditions in your application.
Embrace an intelligent cloud management platform.
Here’s why: Despite configuring CloudWatch and other features of Auto Scaling, you cannot always get everything you need. Further automating various Auto Scaling features using key data-driven insights is the way forward. So, sign-up for an intelligent cloud management platform like Botmetric, which throws key insights to manage AWS Auto Scaling, provides detailed predictive analytics and helps you leapfrog your business towards digital disruption.
Also, do listen to Andre Dufour’s recent keynote on AWS Auto Scaling during the recent 2016 re:Invent, where he reveals that Auto Scaling feature will be available to Amazon EMR (Elastic Map Reduce) service as well along with AWS ECS Container service, and Spot Fleet in regards to dynamic scheduling policies.
It is evident. Automation in every field is upon us. There will soon be a time when we will reach the NoOps state. If you have any questions in regards to AWS Auto Scaling or how you can minimize Ops work with scheduled Auto Scaling or anything about cloud management, just comment below or give us a shout out on Twitter, Facebook, or LinkedIn. We’re all ears! Botmetric Cloud Geeks are here to help.