November 2023 Summaries

Vantage has introduced a new Kubernetes integration that simplifies the process of collecting and uploading cost and efficiency metrics from Kubernetes clusters, replacing the previous OpenCost-based integration. This new Vantage Kubernetes agent offers a single in-cluster deployment, eliminating the need for intermediary services like Prometheus, which previously added complexity and costs. It collects the same metrics as OpenCost, such as CPU, RAM, GPU, and volume costs, but with a more straightforward setup requiring only a Vantage API token. The agent is designed to use fewer resources and is scalable to meet the demands of large clusters. It works across major cloud platforms, including AWS, Azure, and Google Cloud, and provides additional features like Namespace labels, Annotations, and support for Savings Plan discounts. While OpenCost support continues, new users are encouraged to adopt the Vantage agent for its ease of use and efficiency, and those migrating from OpenCost can retain their previous data with assistance from Vantage support.

Amazon EC2 Capacity Blocks are designed to address the growing demand and shortage of GPUs driven by the rise of AI and machine learning applications. These Capacity Blocks allow users to reserve P5 instances with NVIDIA H100 Tensor Core GPUs for short-term needs, up to 14 days, offering a solution particularly beneficial for startups and research organizations that require GPUs for tasks like ML model training and generative AI applications without committing to long-term contracts. Although currently limited to the AWS US East (Ohio) Region, these Capacity Blocks provide flexibility in terms of the number of instances and duration, with dynamic pricing based on supply and demand. This approach not only offers cost savings and increased availability but also facilitates the integration of AI into various projects by making powerful computing resources more accessible.

Revamping a company's tagging strategy is crucial for maintaining a robust FinOps practice and ensuring effective cost management across cloud resources. Tags, which are key/value pairs applied to cloud infrastructure, play a critical role in cost allocation, cloud environment management, performance measurement, reliability, and security. A future-proof tagging strategy should consider compatibility across major cloud providers like AWS, Azure, and GCP, ensuring that tag keys and values are clear and consistent. The strategy should also balance the use of required, optional, and team-specific tags to provide visibility without disrupting workflows. Enforcing a tagging policy can be achieved through either remediation, which involves correcting existing inconsistencies, or enforcement, which ensures tagging compliance at resource creation. The success of implementing a new tagging strategy heavily relies on the support of senior leadership, clear communication, and thorough documentation to engage all stakeholders effectively.

Amazon's introduction of the EC2 M7i-Flex instances, powered by 4th Generation Intel Xeon Scalable processors, represents a significant advancement in cloud computing, offering superior performance and cost-effectiveness compared to the previous M6i instances. These instances are being considered as potential replacements for the T series due to their similar use cases, enhanced performance, and lower costs, although the T4g instances can still offer savings without Unlimited mode enabled. M7i-Flex instances provide a 19% better price performance than the M6i, while T4g instances, running on AWS Graviton2 processors, offer 40% better price performance than T3 instances, making them the cheapest and highest CPU performers of the T series. The decision between M7i-Flex and T4g instances largely depends on the specific use case, as M7i-Flex is better for sustained high CPU utilization, whereas T4g is more cost-effective for workloads that remain under the baseline and avoid Unlimited mode charges. Ultimately, the choice hinges on balancing performance needs with cost considerations, as T4g instances may become more expensive than M7i-Flex when Unlimited mode is used extensively.

Vantage has introduced a new feature for Enterprise customers that allows for automatic team assignment based on existing groups in Single Sign-On (SSO) systems. This enhancement enables organizations to effortlessly mirror their SSO groups as teams within Vantage without additional manual setup, streamlining team management and enhancing efficiency. The feature relies on matching team names in Vantage with those in the identity provider (IdP), using case-sensitive identifiers, and offers the option for custom mappings if necessary. Users are assigned the default user role upon automatic placement into teams, and updates occur at every login. The new functionality is available at no extra cost as part of the Enterprise subscription, and while it does not support SCIM at this time, it supports various IdPs that can pass a groups attribute in the SSO payload.

Vantage has announced the integration of Confluent costs into its platform, allowing users to view and manage their Confluent expenses alongside other cloud infrastructure costs. This feature enables Vantage customers to add multiple Confluent organizations through the Integrations page, where costs are automatically ingested and visualized, simplifying financial operations by consolidating cost data from various providers like AWS, Azure, and Confluent. Confluent, known for its real-time data streaming services based on Apache Kafka, can have costs that escalate with increased data throughput, making this integration beneficial for users needing comprehensive cost visibility. Utilizing the Confluent Costs API, Vantage can securely access billing data, presenting costs with specific filters for services, cost categories, resources, and organizations, which helps users see combined application deployment costs in one view. This integration is available for all Vantage users, including those on the free tier, and supports multiple organization integrations, with daily data refreshes to ensure up-to-date cost insights.

AWS has begun quietly phasing out the transferability of discounted Reserved Instances (RIs), leading to discussions within the FinOps community about the implications of this change. Discounted RIs are those with additional customer-specific discounts beyond the standard rates, and AWS's recent move to stop their transfer could be a response to potential misuse by some vendors who exploit these discounts for profit, leading to what is termed "Reserved Instance Contagion." This practice allowed vendors to increase their revenue by transferring discounted RIs across customers, thus spreading bespoke discounts beyond their intended recipients. The policy shift hints at AWS's possible long-term intention to deprecate EC2 Reserved Instances in favor of more flexible options like Savings Plans, which do not allow for such transfers but provide modern compute savings solutions. Vantage, a FinOps platform unaffected by this change due to its primary revenue from core subscriptions, has audited its users to ensure no exploitation has occurred, reinforcing its commitment to ethical practices and encouraging the use of Savings Plans over RIs.

Inactive VPNs, which have both primary and secondary tunnels down, can accumulate significant costs without providing any operational value, with AWS Site-to-Site VPNs costing five cents an hour per VPN, translating to $36 monthly or $432 annually. For large companies with multiple inactive VPNs, these costs can quickly escalate; for instance, 100 inactive VPNs can cost $3,600 monthly and $43,200 annually. VPNs consist of primary and secondary tunnels, where the primary is the main data transmission pathway, and the secondary acts as a failover. Both tunnels failing means the VPN cannot operate until restored, leading to hidden expenses. To mitigate these costs, it is crucial to routinely check the status of VPN tunnels through methods like API calls, AWS Management Console, CloudWatch, VPN logs, and connectivity tests. Implementing proactive measures such as CloudWatch alerts and regular audits can help ensure that VPN infrastructure is cost-effective and functional, preventing unnecessary charges.

Danielle Vansia's blog post explores the balance between cost and performance when using views in Snowflake, particularly focusing on the differences between non-materialized and materialized views. Non-materialized views, simply referred to as views by Snowflake, do not store data themselves but offer a dynamic way to manage data access and simplify complex queries, while materialized views store pre-computed results to enhance query performance, especially in complex or large datasets. However, materialized views incur additional storage costs, making it essential for data engineers and analysts to assess whether the performance benefits justify the expenses. The article provides practical examples, such as using a non-materialized view for real-time video watch statistics and a materialized view for top-rated videos, to illustrate appropriate use cases. It also suggests optimization strategies, including employing tools like Snowflake's Query History and Search Optimization Service, to enhance data warehouse performance and cost-efficiency. Ultimately, the article emphasizes the importance of carefully considering specific use cases to maintain a cost-effective and efficient data warehouse.

BigQuery is a significant contributor to Google Cloud costs, prompting users to seek ways to optimize expensive queries. Leveraging labels on query jobs offers a practical solution by allowing users to filter and track spending on a per-query basis, providing insights into which queries require optimization. BigQuery pricing primarily consists of compute and storage costs, with the challenge being the lack of a direct method to view query-specific costs within the native interface. Labels can be applied using the bq command-line tool or API, facilitating cost management despite limitations such as the inability to apply them retroactively or the tedious nature of manual labeling. By using labels, users can efficiently identify costly queries and take steps to reduce unnecessary expenses, contributing to effective cost control and resource utilization in BigQuery environments.