July 2023 Summaries

The cloud computing industry saw significant news in March 2021, including a major fire at OVHcloud's Strasbourg data center that damaged two out of four data centers and caused website and email outages for clients. Chinese hackers exploited a loophole in Microsoft Exchange, affecting over 30,000 companies running their own copy of the software. Cloud identity startup Auth0 was acquired by Okta for $6.5 billion. DigitalOcean raised $775 million in an IPO and went public on the NYSE. AWS expanded its presence in Japan with a new region and launched a cost-cutting storage class called One Zone. The FinOps Foundation released a report highlighting the challenges of controlling cloud costs, while Flexera's 2021 State of the Cloud Report showed businesses are increasing their spending on cloud services and starting to use automation for cost control. CAST AI also announced the launch of its Command Line Interface (CLI) and a status page for users, as well as an EKS optimization tool in development.

The ever-growing volumes of data highlight the importance of protecting company assets in the cloud. A comprehensive cloud backup strategy is key to securing data, minimizing downtime risks, and enabling rapid recovery. Assessing application requirements, matching them with suitable solutions, and regularly testing backups are crucial steps in creating a robust backup and recovery plan. The five steps outlined above - identifying app needs, evaluating cloud provider solutions, matchup of requirements and solutions, selecting third-party backup solutions, and deciding on success metrics - can help streamline the process of building a cloud backup strategy.

When managing Kubernetes workloads, pod disruption budgets (PDBs) are crucial for ensuring high availability during voluntary and involuntary disruptions. By setting the minimum number of pods an application needs to function smoothly during disruptions, PDBs provide a safety net, minimizing downtime and ensuring continuity of services. To create a PDB, one must carefully consider the application's availability needs, define appropriate selectors, and apply the YAML file using kubectl commands. Effective use of PDBs involves checking its status, simulating disruptions, and making necessary updates or deletions. Best practices include understanding applications, using selectors wisely, opting for percentage-based disruption budgets, combining with higher-level concepts, accounting for involuntary disruptions, monitoring PDBs regularly, and reviewing and revising them periodically. Real-world examples of PDBs in action include e-commerce platforms, financial services, telecommunications industry, SaaS platforms, and healthcare industry, where they help balance system maintenance with user experience and ensure constant availability. By using pod disruption budgets, developers can optimize Kubernetes resource limits and boost their cluster's performance and cost-efficiency.

Kubernetes Lens is an open-source integrated development environment (IDE) for Kubernetes that simplifies K8s management by letting cloud-native developers manage and monitor clusters in real-time, providing a user-friendly graphical interface to deploy and manage clusters directly from the console. It offers features such as automatic cluster access, discovering local kubeconfig files, managing multiple clusters across diverse platforms, and automatically installing Prometheus instances for metrics and visualizations. Lens also integrates with various Kubernetes tools, provides extensions for customizing its functionality, and enables collaboration through Spaces, allowing teams to organize and access their clusters from anywhere.

KEDA (Kubernetes Event-Driven Autoscaling) is an open-source solution that helps provide a more responsive and dynamic scaling solution for batch jobs, enabling real-time adjustments to match workload demand. It slashes costs by allowing applications to scale down to zero when they are not in use. KEDA can scale the number of job replicas based on messages in Kafka queues, reducing the time it takes to respond to changes in demand. Spot instances offer significant cost savings of up to 90% off on-demand prices, but automation solutions that deliver features like automated provisioning and termination, falling back to on-demand during spot droughts, and partial use of spot instances are crucial for efficient cost optimization. Reentrant batch processing with checkpointing functionality and comprehensive automated testing ensure robust and reliable processing in distributed cloud-native environments. Thin messaging strategies, where messages include a reference or pointer to the actual data stored in scalable object storage, reduce network traffic and increase throughput, making batch jobs easier to scale in phases. Automation platforms designed for Kubernetes can handle autoscaling and spot instances, maximizing cost efficiency while maintaining seamless scaling as workload demands increase.

Cluster Autoscaler is an autoscaling mechanism in Kubernetes that dynamically scales the number of nodes in a cluster based on current utilization, minimizing cloud waste and maximizing ROI from every dollar spent on cloud services. It operates at the infrastructure level, requiring permission to add and delete virtual machines, and should be configured with airtight security for credentials. Cluster Autoscaler works by checking for unschedulable pods, calculating whether consolidating them on fewer nodes is possible, and adding or removing nodes as needed. However, it has limitations, such as not considering CPU or memory utilization in its decision-making process, which can lead to wasted capacity and low utilization efficiency. An alternative to the Cluster Autoscaler Helm Chart is CAST AI autoscaler, a managed service that chooses instance types based on preferences and cloud provider pricing and availability, offering easier configuration and scalability management without performance issues.

Topology spread constraints in Kubernetes provide a fine-grained control over the distribution of pods across failure domains, ensuring high availability and fault tolerance. They can be used to distribute workloads evenly across nodes, regions, and zones, mitigating the risk of node failures affecting the workload. By configuring topology spread constraints, users can ensure that their workloads stay online even if there is an outage or hardware failure in one zone. These constraints also support rolling update workloads and scaling replicas smoothly. They are more flexible than Pod Affinity/Anti-affinity and can be used alongside other scheduling policies like Node Selector or taints.

Spot instances offer significant cost savings, up to 90% lower than on-demand rates, but are often avoided due to concerns about fault tolerance and availability. Automation plays a crucial role in enabling teams to run fault-tolerant clusters on spot instances, reducing cloud costs for Kubernetes users. The key challenges of spot instances include the brief window of time available to replace a reclaimed instance, bidding process risks, and provider capacity limitations. Automation tools can mitigate these challenges by automating spot instance provisioning and termination, providing fallback to on-demand resources when needed, and allowing partial utilization of spot instances with configuration control. By embracing automation, teams can build fault tolerance into their clusters while reducing costs, with CAST AI clients saving an average of 63% on their Kubernetes bills.

The GPU market has been significantly impacted by the recent boom in general artificial intelligence (AI). Despite this, 80% of security professionals remain confident in their existing tooling and processes, even though half have reported a data breach. AWS introduced a lifetime account limit for customers using the Reserved Instance Marketplace. These are some key points from the cloud industry over the past month.