Of late, the cloud has become popular for its ability to deliver agility, scalability, better performance, and cost savings. But ironically, cloud spending has become one of the biggest cost centres for many businesses. Gartner estimates a typical enterprise may overspend on cloud infrastructure by more than 70%. And over 95% of enterprises incur such avoidable cloud-related costs.

Migrating workloads to the cloud takes care of agility and scalability. But performance improvements and cost savings do not realise automatically. These benefits come only through optimising workloads. 

Here is how to optimise cloud workloads and join the small group of smart companies that realise the benefits of the cloud in a cost-optimal way.

1. Select appropriate instance types.

The first step towards cloud performance optimisation is understanding the enterprise workload requirements. Assess resource utilisation of applications and infrastructure components and identify appropriate cloud instances. 

• Compare cloud offerings. Cloud providers offer several instance types, each with varying cost and performance capabilities. Evaluate the specifications and pricing models of these instances. Zero in on the most suitable option depending on the enterprise workload. Consider CPU, memory, storage, and network performance to strike the right balance.
• Use reserved instances when appropriate. Enterprises that subscribe to reserved instances for one to three years save up to 80% of the costs compared to on-demand instances. Such reserved instances suit consistent and predictable workloads.
• Consider cloud storage tiers. Some cloud providers offer hot, warm, and cold storage tiers with varying performance and cost. Store frequently accessed data in higher-performance tiers and less frequently accessed or archival data in lower-cost tiers. Such tiering reduces storage costs while still ensuring ready data availability.
• Leverage spot instances. Cloud providers such as AWS offer spot instances. Google Cloud offers preemptible VMs. These resources cost less than regular instances but come with the risk of interruption at short notice. Spot instances remain a cost-attractive option to host fault-tolerant and non-critical workloads. For instance, batch jobs can tolerate disruption without any implication for performance. Amazon EC2 spot instances, for instance, offer unused EC2 capacity in the AWS Cloud at discounts of up to 90% compared to on-demand Instance prices. But AWS can also interrupt these spot instances with a two-minute notification. 
• Use tools that make provisioning easy. For instance, Amazon EKS, a fully managed Kubernetes service, provisions resources with less operational effort than self-managed node groups. Amazon EKS facilitates running Kubernetes on AWS without installing, operating, and maintaining a Kubernetes control plane independently.

2. Track and adjust resources in real-time

Tracking resources in real time optimises performance and ensures cost-effectiveness.

Cloud monitoring tools offer insights into resource utilisation, performance bottlenecks, and cost patterns. 

Identify critical application performance metrics to set minimum performance baselines. Commonly-used metrics include response time, throughput, and scalability. Track such performance metrics continually, and apply analytical tools to get actionable information. Use such insights to identify over-provisioned or underutilised resources. Adjust resource allocations to match workload demands.

Automate resource deployment to resize resources appropriately. Automation enables system admins to make quick changes. It furthers cloud cost optimisation by 

• Shutting down workloads after required hours
• Rightsizing instances
• Cutting down unneeded resources 
• Enabling experimentation with different instance types, sizes, and configurations. For instance, system admins could try a high-end server for some time to see how the applications meet the demand.

3. Utilise auto-scaling

Enterprises have diverse workloads, and their computing and memory requirements vary with time. 

But cloud scalability often comes at a heavy price. Many cloud users incur sizable costs to provision for unexpected traffic spikes and load fluctuations. Serverless computing overcomes scaling issues but may incur runaway costs. 

Auto-scaling adjusts resources based on workload demand. It makes available resources during peak periods and sheds resources during periods of lower demand. Scaling down elastically removes compute resources without an impact on the underlying processes.

Popular services such as Amazon EKS and Amazon Elastic Compute Cloud (EC2) scale resources to meet workload demands. 

To keep costs in check when autoscaling:

• Set up auto scaling parameters with cost in mind. For example, put performance limits on lower-priority workloads that do not need extensive scale. 
• Configure auto scaling settings to use the minimum resources to meet demand. 
• Use queuing and caching to accommodate unexpected traffic spikes instead of paying for idle capacity.

4. Optimise data transfers 

Cloud pricing is dynamic and depends on peak times and compute demands. As such, many users move workloads to and from geographic regions with low demand for lower processing costs. The major prerequisite is the ability of common storage services to support each location. 

Many enterprises also use data centres in multiple zones to limit the impact of potential outages in a single data centre. 

Enterprises also adopt a multi-cloud strategy. They may perform non-sensitive tasks on different cloud platforms that offer cost advantages. 

But data transfer traffic between regions, availability zones, and different services is costly. Also, database replication between regions causes latency issues.

To keep costs in check when moving data between cloud:

• Apply content compression and optimise data transfer protocols to minimise bandwidth requirements.
• Implement data caching. Caching stores data closer to the end users and minimises data transfer across zones. In-memory caching services and content delivery networks reduce the load on the infrastructure.
• Make use of specialised tools offered by cloud providers. For instance, AWS DataSync and AWS Transfer Family provide efficient and cost-effective data transfer options across regions and zones.

5. Embrace the edge

In edge computing, data processing occurs in distributed infrastructure deployed near the data source. The setup improves performance by overcoming bandwidth requirements and latency issues. Data is stored locally, and only the processed insights are transmitted to centralised cloud servers. Bandwidth needs to be reduced drastically.

The edge offers a host of advantages for enterprises. Enterprises gain direct savings on bandwidth and storage costs. The improved performance also reduces the cost of latency-sensitive applications.

6. Review licensing costs

Have a look at licensing costs. Migrating workloads to the cloud often leads to the wastage of traditional software licences, such as MS Windows Server. The bring-your-own-licence (BYOL) approach allows using existing licences without additional fees. 

• Check the possibility of BYOL.
• Understand the cost implications of licences. 

Cloud success depends on ensuring optimal performance without unnecessary cost. The cloud-first approach is a rage. But blind adoption of the concept is often counterintuitive. Enterprises need to strike the right balance between cloud performance and cost. They need the right cost management strategies to be agile and resilient. Optimising the cloud infrastructure is a continuous process of evaluation and making refinement.