Welcome to issue #443 March 24th, 2025

News

Google + Wiz: Strengthening Multicloud Security - Google Cloud announced the acquisition of Wiz, a cloud security platform provider, to enhance its multicloud security offerings.

Announcing BigQuery repositories: Git-based collaboration in BigQuery Studio - BigQuery repositories, a new feature in BigQuery Studio, enables data teams to collaborate on code stored in Git repositories. It provides a comprehensive set of features to integrate Git workflows directly into the BigQuery environment, including setting up new repositories, connecting to remote Git hosts, editing code, and performing Git operations.

Cloud Composer 3: The next generation of data pipeline orchestration - Cloud Composer 3, the latest version of Google's fully managed Apache Airflow service, simplifies data pipeline management and boosts efficiency. It offers simplified networking, evergreen versioning, hidden infrastructure, enhanced performance, per-task CPU and memory control, and robust security features.

Introducing protection summary, a new Google Cloud Backup and DR feature - Google Cloud Backup and DR protection summary is a new feature that provides a centralized view of your backup configurations, helps you identify gaps in your data protection, and empowers you to take action to improve your resilience. It allows you to quickly discover resources that lack any backup configuration and take immediate steps to protect them.

Using RDMA over Converged Ethernet networking for AI on Google Cloud - Google Cloud introduces RDMA over Converged Ethernet networking (RoCE v2) for high-performance AI workloads. RoCE v2 enables direct data exchange between systems without involving the OS, networking stack, and CPU, resulting in faster processing times.

Accelerate AI/ML workloads using Cloud Storage hierarchical namespace - Cloud Storage hierarchical namespace (HNS) is a new feature that can improve the performance and efficiency of AI/ML workloads. HNS provides a "folder" resource type and APIs that are optimized for filesystem semantics, resulting in faster and more reliable checkpointing. It also has an optimized storage layout that handles higher queries per second (QPS) of reads and writes.

Building AI agents with Gen AI Toolbox for Databases and Dgraph - Gen AI Toolbox for Databases now supports Dgraph, a fully open-source, built-for-scale graph database for AI apps. The integration simplifies configuration and development, provides production-ready infrastructure, and offers enterprise-grade observability. Dgraph's real-time performance, horizontal scalability, AI-native primitives, and flexible data modeling make it ideal for building sophisticated gen AI tools.

Vertex AI Search and Generative AI (with Gemini) achieve FedRAMP High - Google Vertex AI Search and Generative AI (with Gemini) have achieved FedRAMP High authorization, ensuring robust security for mission-critical applications in the public sector. Vertex AI Search enables multimodal Google-quality search across external, internal, and proprietary data, enhancing discoverability and transparency.

Vector similarity search for Cloud SQL for MySQL is now GA - Cloud SQL for MySQL now supports vector storage and similarity search, allowing you to transform your MySQL databases in place to integrate generative AI capabilities without a specialized vector database. You can add a new column to your existing table, load in your vector embeddings, and perform k-nearest neighbors (kNN) and approximate nearest neighbors (ANN) searches.

Build richer gen AI experiences using model endpoint management - Model endpoint management is now available on AlloyDB, AlloyDB Omni, and Cloud SQL for PostgreSQL. It allows developers to build new experiences using SQL and provides a flexible interface to call generative AI models running anywhere — right from the database. You can generate embeddings inside the database, perform quality control on your vector search, and analyze sentiment in the database, making it easier to monitor results.

Elevate your workload resilience in console with data protection for Compute Engine - Google Cloud introduces a streamlined data protection experience for Compute Engine workloads, simplifying backup and disaster recovery (DR) management. The new Data Protection tab in the VM creation interface consolidates backup and continuous data replication options, allowing users to configure data protection settings when creating VMs.

Gen AI Toolbox for Databases announces LlamaIndex integration - LlamaIndex, a leading framework for building knowledge-driven and agentic systems, has integrated with Gen AI Toolbox for Databases, an open-source server that streamlines the development and management of sophisticated generative AI tools that can connect to databases.

Google Cloud at GTC: A4 VMs now generally available, A4X VMs in preview - Google Cloud is showcasing its latest advancements in AI and accelerated computing at NVIDIA's GTC AI Conference in San Jose, California. The company announced the general availability of A4 VMs and the preview of A4X VMs, powered by NVIDIA's HGX B200 and GB200 NVL72 GPUs, respectively.

Accelerating AI in healthcare using NVIDIA BioNeMo Framework and Blueprints on GKE - NVIDIA BioNeMo, an open-source collection of models for medical and pharmaceutical research, combined with Google Kubernetes Engine (GKE), offers a powerful solution for accelerating AI in healthcare. BioNeMo provides domain-specific optimization, GPU-accelerated performance, and comprehensive open-source resources for building and deploying biomolecular AI solutions.

Gemini in Workspace apps and the Gemini app are first to achieve FedRAMP High authorization - Gemini in Workspace apps and the Gemini app are the first generative AI assistants for productivity and collaboration suites to achieve FedRAMP High authorization. This enables government agencies to confidently deploy Gemini's advanced AI capabilities within the stringent FedRAMP High security framework. Gemini can automate tasks, optimize processes, and free up government employees to focus on higher-value activities.

Co-op mode: New partners driving the future of gaming with AI - Google Cloud partners with nunu.ai, ElevenLabs, and Common Sense Machines to enhance game development with AI.

Strengthening Google Developer Experts community with Google Cloud Champion Innovators - The Google Developer Experts (GDE) program is merging with the Google Cloud Champion Innovators (Champions) program, creating a single, powerful community of over 1,400 experts. This integration aims to amplify the impact of passionate experts, foster collaboration, and provide a stronger voice for developers within Google and the industry.

---

Webinar

Cut Kubernetes Costs: 4 Key Questions Every CFO Should Ask Their Engineering Leader

As a trusted Google Cloud partner of over a decade, DoiT delivers the only end-to-end FinOps platform that goes beyond cost optimization to drive reliability, performance, and security. In this webinar, we’ll break down the four most common drivers of waste in containerized workloads and give you the exact questions to ask your engineering leader to take action. You’ll walk away with a CFO-friendly framework for tackling Kubernetes cost overruns, improving cloud efficiency, and strengthening collaboration between Finance and Engineering. Register

---

Articles, Tutorials

Infrastructure, Networking, Security, Kubernetes

BitM Up! Session Stealing in Seconds Using the Browser-in-the-Middle Technique - BitM (Browser-in-the-Middle) attacks allow attackers to quickly compromise sessions across various web applications, bypassing even sophisticated social engineering tactics that target session tokens. While multi-factor authentication (MFA) remains crucial, organizations must implement robust defenses such as hardware-based MFA, client certificates, and FIDO2 to counter these threats.

Harvesting hardware: Our approach to carbon-aware fleet deployment - Google's hardware harvesting program prioritizes the reuse of existing hardware when deploying new machines or components, reducing carbon emissions, conserving resources, and minimizing waste. By reusing over 293,000 components in 2024, the program saved carbon emissions and reduced costs.

Cloud CISO Perspectives: 5 tips for secure AI success - Google Cloud CISO Phil Venables and Royal Hansen, VP of Engineering for Privacy, Safety, and Security, discuss how new AI Protection capabilities in Security Command Center fit in with Google Cloud's overall approach to securing AI. They offer five tips to help organizations ensure secure AI success: implement strong AI governance, use good data, control access, address inherited vulnerabilities, and mitigate internal AI tool risks.

Oracle Database Operator for Kubernetes on Google Cloud - This article describes the steps for installing the Oracle Database Operator for Kubernetes (OraOperator) on the Google Kubernetes Engine (GKE) and provisioning a Single Instance Database (SIDB).

Serving Multiple Load Balancers with a Single GCE Instance on Unique Ports - This guide shows how to serve multiple ports from a single GCE instance by configuring multiple HTTP load balancers, each routing traffic to a different port.

Migrating to cgroups v2 in GKE: A Tale of Clusters and Containers - Migrating to cgroups v2 in GKE involves upgrading the cluster to use the latest version of cgroups, which provides better resource management and security.

Anycast IP location issue - Why Your Google Cloud Global Load Balancer IP Shows Up in the US (and Why It’s Misleading).

App Development, Serverless, Databases, DevOps

Enhancing multiplayer games with gen AI and cloud gaming: a technical deep dive - This blog post showcases how a sample game can be enhanced with generative AI, enabling developers to enrich their games with AI-driven features such as interactive user guidance, real-time chat capabilities, and personalized summaries of user progress. The potential for gen AI in gaming is enormous, and the tools are now more accessible than ever.

JetStream for GCE Disaster Recovery Orchestration: Protect and manage your critical workloads - JetStream for GCE Disaster Recovery Orchestration, available on Google Cloud Marketplace, offers near-zero recovery objectives and orchestrates protection, failover, and failback operations. It uses Google Cloud's Asynchronous Replication for low RPO and RTO, and supports consistency groups for multi-disk applications.

A framework for adopting Gemini Code Assist and measuring its impact - Gemini Code Assist is an AI-powered tool that helps developers write better code. It can be adopted by following a four-step framework: Adoption, Trust, Acceleration, and Impact.

AlloyDB for PostgreSQL: Two years of innovation and industry leadership - AlloyDB for PostgreSQL, a fully managed database service built on Google Cloud's infrastructure, offers exceptional performance, scalability, and analytical capabilities. With its disaggregated compute and storage architecture, optimized database kernel, and read pools, AlloyDB delivers fast transactional processing and massive horizontal scaling for read-intensive workloads.

Leveraging Google Cloud Functions to Create a Powerful Fivetran Connector to integrate custom APIs - Fivetran, a cloud-based ETL service, simplifies data ingestion from various sources, including custom APIs, to your data warehouse. By leveraging Google Cloud Functions as connectors, you can automate data extraction and transformation from your custom API, eliminating manual coding and maintenance.

FinOps in Action: Slash Google Cloud Costs After Lift & Shift with Automated VM Scheduling - Use Cloud Workflows and Cloud Build to remove Idle VM Costs on Google Cloud.

Big Data, Analytics, ML&AI

The AI lens: How Arpeely uses multimodality and BigQuery to revolutionize AdTech - Arpeely uses Google Cloud AI to revolutionize AdTech by leveraging multimodality and BigQuery. Their AI algorithms analyze massive datasets to identify high-value customers, using multimodal AI to analyze web page screenshots and extract visual information. BigQuery helps them crunch numbers, analyze user behavior, and generate insights from over 25 petabytes of compressed data.

Build GraphRAG applications using Spanner Graph and LangChain - Spanner Graph and LangChain streamline GraphRAG development by combining Spanner Graph's enterprise-grade reliability, scalability, and distributed graph processing with LangChain's versatile tools. This enables rapid prototyping of intelligent applications and unlocks valuable data insights. GraphRAG outperforms conventional RAG by providing richer, more informative answers through graph traversals and context retrieval.

Mastering secure AI on Google Cloud, a practical guide for enterprises - Learn how to implement secure AI solutions, addressing challenges around data privacy and security. Discover the four cornerstones of a secure AI platform: infrastructure, data, security, and responsible AI.

Five tips and tricks to improve your AI workloads - Optimize your AI workloads on Google Cloud Platform and reduce costs. Explore five tips to improve workflow, including choosing the right platform, optimizing inference container startup time, selecting suitable storage options, securing resources with Dynamic Workload Scheduler and Future Reservations, and using custom disk images for faster setup.

BigQuery Cost Optimization with slot management - This article covers essential cost-saving techniques, including optimizing slot usage, query performance, and proactive monitoring.

Build a vector search application in Python using BigFrames - Vector search, powered by generative AI, is the key to unlocking valuable insights from unstructured data. BigFrames, an open-source Python library, empowers data scientists and developers to leverage familiar Python tools like Pandas and Scikit-learn for vector search within BigQuery.

How to Implement Hybrid Search for RAG with BigQuery - This article discusses how to implement hybrid search for RAG (Retrieval Augmented Generation) with BigQuery. It explains the shortcomings of traditional RAG approaches that rely solely on vector search and introduces the concept of hybrid search, which combines keyword and vector searches.

BigQuery’s Ridiculous Pricing Model Cost Us $10,000 in Just 22 Seconds!!! - Yes, you read that right. $10,000 burned in less than half a minute.

Oracle to BigQuery Migration: The Challenges, The Fixes, and The Wins - The article discusses the challenges, fixes, and wins experienced during the migration of an entire data warehouse from Oracle to BigQuery. The migration involved over 100 terabytes of data, 5000 tables, and complex transformation requirements. Despite encountering data type mismatches, SQL rewrites, performance issues, and hidden costs, the team successfully completed the migration and achieved a modern, scalable, and cost-effective cloud solution.

Various

The three pillars of data-driven government - Google's experiences in the public sector have led to the development of a framework for governments to define and refine their AI and data access strategy. This framework is based on three core pillars: defining a government Data Access Platform (DAP), assembling a central DAP empowerment team, and empowering ministries and government agencies to contribute to and benefit from the DAP.

Google Cloud Next 25 Partner Summit: Session guide for partners - Google Cloud Next '25 Partner Summit is an opportunity for partners to hear from Google Cloud leaders on upcoming developments and strategies for 2025.

An inside look into Google's AI innovations: AI Luminaries at Cloud Next - Google Cloud Next, an upcoming conference, will feature AI Luminaries programming, where top researchers, scientists, and technology leaders will engage with Google experts in critical AI domains.

Slides, Videos, Audio

Security Podcast - #215 Threat Modeling at Google: From Basics to AI-powered Magic.

Releases

Access Context Manager - Access Context Manager now supports custom organization policies.

Agent Assist - Build your own Gen AI Assist is available in preview. Agent Assist offers Vertex extensions for Build your own assist (BYOA) in preview.

AlloyDB - Performing an in-place major version upgrade of your AlloyDB cluster is generally available (GA). You can use a query recall evaluator (Preview) to find the recall for a vector query for a given configuration, and to tune your parameters to achieve the desired vector query recall results. Outbound connectivity for Private Service Connect-enabled AlloyDB clusters is generally available (GA).

Anthos clusters on VMware - Google Distributed Cloud (software only) for VMware 1.31.300-gke.81 is now available for download. The 1.31.300-gke.81 release includes many vulnerability fixes.

AppEngine Standard - Container Registry is now shut down.

BigQuery - BigQuery workflows have been renamed to BigQuery pipelines in the Google Cloud console. You can now use repositories and workspaces in BigQuery to perform version control. You can now create remote models in BigQuery ML based on the Anthropic Claude model in Vertex AI. You can now use EXPORT DATA statements to reverse ETL BigQuery data to Spanner. You can now create an external dataset in BigQuery that links to an existing database in Spanner. You can now use the TYPEOF function to determine the data type of an expression.

Cloud Build - In the filtering toolbar of the Triggers page, you can now filter by trigger repository and region. You can now specify, in your build config file, a custom Pub/Sub topic for build notifications.

Chronicle - Statistics and aggregations in UDM search using YARA-L 2.0 You can now run statistical queries on UDM events and group the results for analysis using YARA-L 2.0.

Chronicle Security Operations - The following parser documentation is now available: Collect AWS Config logs Collect AWS Elastic Load Balancing logs Collect AWS Route 53 logs Collect AWS S3 server access logs Collect AWS WAF logs Collect Azure Application Gateway logs Collect Carbon Black App Control logs Collect Carbon Black EDR logs Collect Delinea Secret Server logs Collect Radware WAF logs Collect AWS Aurora logs Collect AWS CloudWatch logs Collect AWS Control Tower logs Collect AWS Elastic MapReduce logs Collect AWS Key Management Service logs Collect AWS Macie logs Collect AWS Network Firewall logs Collect AWS Security Hub logs Collect AWS Session Manager logs Collect Zscaler DLP logs Collect Zscaler Tunnel logs Collect Zscaler VPN logs Collect Zscaler ZPA Audit logs Collect Zscaler ZPA logs Collect Zscaler CASB logs Collect Azure AD Sign-In logs Collect Azure API Management logs Collect Azure APP Service logs Collect Azure Firewall logs Collect Azure VPN logs Collect AWS VPN logs Collect Azure Storage Audit logs Collect Azure WAF logs Collect Cloud IoT logs Collect Cloud Run logs Collect Cloud Compute logs Collect CrowdStrike Falcon Stream logs Collect SentinelOne Deep Visibility logs Collect Cloud VPC Flow Logs Collect Cloud Compute context logs Collect Cloud Intrusion Detection System (Cloud IDS) logs Collect Cloud Next Generation Firewall Enterprise logs Collect Cloud Storage context logs Collect Cloud Identity and Access Management (IAM) Analysis logs Collect Cloud Identity Devices logs Collect Cloud Identity Device Users logs Collect Cloud Security Command Center Error logs Collect Cloud Security Command Center Observation logs Collect Cloud Security Command Center Posture Violation logs Collect Cloud Security Command Center Toxic Combination logs Collect Cloud Security Command Center Unspecified logs Collect Cloud Secure Web Proxy logs. Statistics and aggregations in UDM search using YARA-L 2.0 You can now run statistical queries on UDM events and group the results for analysis using YARA-L 2.0.

Chronicle SOAR - Release 6.3.40 is being rolled out to the first wave of regions as listed here. Theme enhancement for SOAR platform The header and left hand navigation menu now fully reflect the selected theme. Release 6.3.39 is now available for all regions.

Cloud Composer - After April 15, 2025 the database retention policy feature will be enabled by default in newly created Cloud Composer 3 environments. The issue with Cloud Composer 2 upgrade operations is now resolved.

Compute Engine - Generally available: Resource-based committed use discounts (CUDs) are available for licenses of RHEL operating system images. Generally available: The A4 accelerator-optimized machine type is now generally available.

Container Registry - Container Registry is shut down and writing images to Container Registry is unavailable.

Data Catalog - Data Catalog is available in the europe-north2 (Stockholm) region.

Data Fusion - Cloud Data Fusion version 6.11.0 is available in Preview. You can view instance metrics and pipeline metrics in Cloud Monitoring and in the dashboard provided by Cloud Data Fusion. You can view instance logs and pipeline logs in Cloud Logging, and in the dashboard provided by Cloud Data Fusion. When a pipeline run fails, you can retrieve detailed error information on the pipeline details page of the Cloud Data Fusion web interface. Cloud Data Fusion 6.11.0 offers high availability with reduced upgrade downtime. Changes in Cloud Data Fusion 6.11.0: To create ephemeral clusters, Cloud Data Fusion uses the Dataproc 2.2 image by default. Fixed in Cloud Data Fusion 6.11.0: Fixed a null pointer exception in the BigQuery multi-sink plugin when used without a reference name (PLUGIN-1843). A soft limit of 2 MB for pipeline JSON size is introduced in 6.11.0. The following APIs for searching and querying metrics are deprecated in 6.11.0: POST v3/metrics/query POST v3/metrics/search. The following APIs for downloading system service and pipeline run logs are deprecated in 6.11.0: GET /v3/namespaces//apps////logs GET /v3/system/services//logs. The ability to retrieve all applications without pagination using the GET /v3/namespaces//apps endpoint is deprecated in 6.11.0.

Dataplex - Dataplex and data lineage are available in the europe-north2 (Stockholm) region.

Dataproc Serverless - New Dataproc on Compute Engine subminor image versions: 2.0.136-debian10, 2.0.136-rocky8, 2.0.136-ubuntu18 2.1.84-debian11, 2.1.84-rocky8, 2.1.84-ubuntu20, 2.1.84-ubuntu20-arm 2.2.50-debian12, 2.2.50-rocky9, 2.2.50-ubuntu22. Dataproc on Compute Engine: Spark upgraded to version 3.5.3 in the latest Dataproc image version 2.2. Dataproc on Compute Engine: The latest Dataproc 2.2 image version now supports Spark data lineage.

Dataproc - New Dataproc on Compute Engine subminor image versions: 2.0.136-debian10, 2.0.136-rocky8, 2.0.136-ubuntu18 2.1.84-debian11, 2.1.84-rocky8, 2.1.84-ubuntu20, 2.1.84-ubuntu20-arm 2.2.50-debian12, 2.2.50-rocky9, 2.2.50-ubuntu22. Dataproc on Compute Engine: Spark upgraded to version 3.5.3 in the latest Dataproc image version 2.2. Dataproc on Compute Engine: The latest Dataproc 2.2 image version now supports Spark data lineage.

Datastream - You can now use Secret Manager to securely store authentication resources with Datastream.

Cloud Deploy - Cloud Deploy now uses Skaffold 2.14 as the default Skaffold version, as of March 21, 2025, for all target types.

Eventarc - Eventarc Standard is available in the europe-north2 (Stockholm, Sweden) region.

Gemini - Streamed chat responses are now available in public preview for IntelliJ and VS Code Gemini Code Assist. You can now configure and use custom commands in the inline chat menu and lightbulb menu for VS Code Gemini Code Assist. Fixed an issue with an infinite progress bar while trying to log in to IntelliJ Gemini Code Assist.

Google Kubernetes Engine - In GKE version 1.32.2-gke.1652000 and later, new external LoadBalancer Services use zonal Network Endpoint Group (NEG) backends by default. All GKE clusters now export four new rollup metrics by default at no additional charge. Starting in GKE version 1.32.1-gke.1729000, Autopilot clusters will automatically use the new Performance HPA Profile. (2025-R11) Version updates GKE cluster versions have been updated. On GKE clusters running versions 1.32.2-gke.1182000 to 1.32.2-gke.1297000, Pods using Cloud Storage FUSE CSI driver volumes (persistent or CSI ephemeral) fail to schedule when both of the following are true: The Pods run on the host network (hostNetwork: true) You use a custom sidecar image built with a Cloud Storage FUSE CSI driver version earlier than v1.12.2.

GKE new features - In GKE version 1.32.2-gke.1652000 and later, new external LoadBalancer Services use zonal Network Endpoint Group (NEG) backends by default. All GKE clusters now export four new rollup metrics by default at no additional charge. Starting in GKE version 1.32.1-gke.1729000, Autopilot clusters will automatically use the new Performance HPA Profile.

GKE - (2025-R11) Version updates Note: Your clusters might not have these versions available.

Load Balancing - Google Cloud periodically renews Google-managed certificates by requesting them from certificate authorities (CAs).

Looker - Looker (Google Cloud core) and Looker (original) changes. The following features have been added to Studio in Looker, which is available in preview: If Studio in Looker is disabled and then re-enabled, reports that had been saved within the previous 30 days will still be available.

Cloud Monitoring - The Google-Built OpenTelemetry Collector is now available. You can now enable and disable uptime-checks by using the disabled field in the Cloud Monitoring API.

Resource Manager - You can enforce mandatory tags on resources using custom organization policies.

Security Command Center - The Risk section of the SecOps console has been updated for Security Command Center Enterprise, introducing the following features in Preview: Issues are the most important security risks Security Command Center Enterprise has found in your cloud environments. Cloud Infrastructure Entitlement Management (CIEM) has launched support for the following: Log ingestion from Amazon SQS queues.

Sensitive Data Protection - The CZECHIA_PASSPORT infoType detector is available in all regions.

Service Mesh - Managed Cloud Service Mesh & 1.22.x & 1.23.x & 1.24.x. You can now use custom constraints with Organization Policy to provide more granular control over specific fields for some networksecurity and networkservices resources. Cloud Service Mesh now supports dual-stack, extending IPv6 capability to both proxy-based Envoy and proxyless gRPC. Managed Cloud Service Mesh. If you use the managed Cloud Service Mesh with the ISTIOD control plane implementation, important changes have been made to how and when you'll receive notifications of upcoming modernization.

Cloud Spanner - The default time zone of your Spanner databases can now be set. You can now create an external dataset in BigQuery that links to an existing database in Spanner. You can now use EXPORT DATA statements to reverse ETL BigQuery data to Spanner.

Cloud SQL MySQL - You can now use PITR to restore a Cloud SQL for MySQL instance that isn't available.

Cloud SQL SQL Server - Cloud SQL for SQL Server supports transparent data encryption (TDE) to encrypt data stored in your Cloud SQL for SQL Server instances.

Cloud Storage - Storage Intelligence for Cloud Storage is now generally available (GA). Storage Insights datasets is now generally available (GA). Cross-bucket replication is now generally available (GA). Cloud Storage now offers the DE configurable dual-region code, which can be used when creating a dual-region bucket in europe-west3 (Frankfurt) and europe-west10 (Berlin).

Cloud Text-to-Speech - Chirp 3: HD voices are only available in the global, us, eu, and asia-southeast1 regions.

Vertex AI Workbench - Encrypt your data-in-use by using Confidential Computing.

VMware Engine - VMware Engine is integrated with Google Cloud Essential Contacts for email notifications.

VPC Service Controls - VPC Service Controls now supports custom organization policies.

Workload Manager - Generally available: Workload Manager supports deploying Microsoft SQL Server workloads on Google Cloud. Workload Manager supports the following features when you deploy SAP S/4HANA workloads on Google Cloud: Deploy SAP S/4HANA workloads on X4 instances.