Templatized Configurations: A Double-Edged Sword

AWS offers multiple managed services such as RDS, ElastiCache and many more. Along with the managed services AWS also offers templatized configurations for dev/test or production environments. These templatized configurations provide an easy start and access to best practices for generic usage patterns.

Advantages of Templatized Configurations:

• Ease of Deployment: Templates significantly reduce setup time and efforts by providing pre-configured settings.
• Best Practices: Templates often incorporate recommended best practices and security measures.

However, these templates are often only a great way to start and may not be most appropriate as you scale your cloud environments.

Disadvantages of Templatized Configurations:

• Best for generic use cases: standardized templates are built for a few generic use cases and could work best if your use-case adheres to the intended use case. ?
• Mismatch with application SLAs: your intended RTO, RPO, and availability SLAs are not factored into these templates, they could be under or over-provisioned for your SLAs
• Mismatch for your demand patterns: suboptimal configurations for your demand seasonality could mean wasted spend or below-par performance

?

A Real World Example

We have been working with example.com (company name anonymized), a leading AI-based B2B company. Example.com is cloud native and leverages several AWS managed services. These managed services are often launched with templatized configurations. ?

Example.com has been using ElastiCache as a caching layer for multiple system components. zOpt.ai identified an optimization opportunity for multiple ElastiCache deployments. This promised a potential 30% savings in ElastiCache cluster costs.

How did zOpt.ai identify this optimization?

ElastiCache clusters were deployed using preconfigured templates, 2+ shards with 2+ nodes per shard. Although ElastiCache cluster performance is evaluated on several metrics, let's just focus on two of them: CPU consumption across cluster nodes and Cache hit rate across nodes per shard. Based on these 2 data points, you could determine whether the cluster is underprovisioned, overprovisioned or provisioned appropriately. This analysis needs to be performed over a prolonged period.?

CPU utilization: Analyze CPU utilization of all the nodes in the cluster over 3 3-month period and verify how busy or free these nodes are. In the chart below, you can see the CPU utilization is consistently low, ~5% each across the nodes in the cluster. This means this cluster is very likely over-provisioned for vCPUs.

Cache hit rate: Analyze the cache hit rate for all nodes across the cluster over 3-month period and verify if the nodes are busy, and whether the traffic is distributed across all nodes in the cluster. In the chart below, you can cache hit rate is only observed on cache-0001-001 node while other nodes do not see any activity. This means the cluster is over-provisioned and the traffic isn’t distributed evenly across the cluster.

With zOpt, all such analysis is performed automatically for you and the most optimum configuration is brought out as a recommendation. In this specific case, Example.com was given the optimization recommendation that they could implement the changes using zOpt’s human-vetted automation. Example.com could save 30% of the ElastiCache spend with a few clicks, and without worrying about any adverse impact on their application performance.

How can you avoid this?

• Assess Your Specific Needs: Carefully evaluate your application's requirements before choosing a template.
• Optimize Configurations: Fine-tune parameters like instance size, storage, and network settings to match the demand seasonality of your workload.
• Monitor and Adjust: Continuously monitor resource utilization and make adjustments as needed.

zOpt to the Rescue zOpt, an AI-powered cloud optimization platform, can automatically identify such overprovisioning and inefficient configurations. By analyzing your AWS usage patterns, zOpt provides actionable recommendations to optimize your costs and performance.

Let's work together to maximize the value derived from your AWS investments.

#AWS #cloudoptimization #cloudcosts #cloudnative #devops #cloudsecurity #awscoustamazon #cloudcomputing #cloudmigration #cloudengineering