rodric rabbah

People often ask me about the differences in architecture between Amazon Dynamo (the 2007 SOSP paper), DynamoDB (the AWS serverless NoSQL database), and Aurora DSQL (the AWS serverless SQL databases). I captured the key points on durability, high availability, programming model, and trade-offs in my blog: https://lnkd.in/gkMG_tVD TL;DR: Despite the shared name, Dynamo and DynamoDB have little in common architecturally, and make very different trade-off decisions.

631

8 Comments

Today, at #GoogleCloudNext, we’re announcing significant improvements to Google Kubernetes Engine (GKE) to help platform teams succeed with AI: * Cluster Director for GKE, now generally available, lets you deploy and manage large clusters of accelerated VMs with compute, storage, and networking — all operating as a single unit. * GKE Inference Quickstart, now in public preview, which simplifies the selection of infrastructure and deployment of AI models, while delivering benchmarked performance characteristics. * GKE Inference Gateway, now in public preview, provides intelligent routing and load balancing for AI inference on GKE. * A new container-optimized compute platform is rolling out on GKE Autopilot today, and in Q3, Autopilot’s compute platform will be made available to standard GKE clusters. * Gemini Cloud Assist Investigations, now in private preview, helps with GKE troubleshooting, decreasing the time it takes to understand the root cause and resolve issues. * With a built-in partnership with Anyscale, RayTurbo on GKE will launch later this year to deliver superior GPU/TPU performance, rapid cluster startup, and robust autoscaling. More details in the blog post below... https://lnkd.in/gxuzMJaJ

263

Google Spanner is the battle-tested distributed-SQL database that delivers planet-scale strong consistency and supports Google billion-user applications like YouTube, GMail and Google Ads. Spanner offers multiple data models and query models, including SQL, Graph, Key/Value, Vector and Search. Spanner now also supports tiered storage, as described below. https://lnkd.in/eHEmwTiP

71

2 Comments

Big news for AI workloads on Oracle Cloud Infrastructure: OCI + DDN Infinia are redefining how enterprises scale AI. In recent benchmarks, Infinia on OCI showed linear scalability, ultra-low latency, and GPU utilization near 100% — a huge step forward for running production-scale AI. This means enterprises can: ✅ Break down data silos with unified AI storage ✅ Support exabyte-scale datasets + 100Ks of GPUs ✅ Accelerate real-world use cases from finance to healthcare 👉 Full blog here: https://lnkd.in/emkxJuvE 💬 Curious — what’s your biggest storage bottleneck when scaling AI today? Thanks for publishing Pinkesh Valdria, Dhvani S., Salem Albeiroti Kumaran Rajaram

96

1 Comment

As the complexity of cloud environments grows, so does the need for simplified, unified security. In the second of a two-part series with Frontier Enterprise, Datadog CEO Olivier Pomel shares why integrating security into the entire stack is critical—not just for faster response, but for stronger alignment across engineering, operations, and security teams. Read the full story: https://bit.ly/45kBPDj

83

1 Comment

We're excited to announce new capabilities in Google Cloud #VertexAI Training designed to simplify and accelerate large-scale model development. Key highlights from our latest update: 🔹 Flexible, Self-Healing Infrastructure: Leverage fully managed, resilient Slurm environments with automated failure detection and performance-optimized checkpointing. 🔹 Cost-Effective Scheduling: Utilize Dynamic Workload Scheduler (DWS) for fixed future reservations or flexible on-demand capacity. 🔹 Integrated Frameworks: Access optimized recipes for the full model lifecycle (including SFT and DPO) and seamless integration with NVIDIA NeMo. Whether you're fine-tuning standard models or training massive custom foundational models from scratch, these new features are built to get you to production faster. Read the full announcement to see how organizations like Salesforce and AI Singapore are already leveraging these tools to elevate their models. 👇 https://lnkd.in/gVVq3WCw

185

2 Comments

AI teams need scalable, reproducible, and intelligent data workflows. The solution? Combining lakeFS and Red Hat OpenShift AI. AI projects demand more than advanced models; they need infrastructure that ensures data integrity, accelerates iteration cycles, and supports MLOps best practices. Many teams struggle to achieve this at scale… Here’s where lakeFS + OpenShift AI step in. lakeFS provides Git-like versioning for datasets, enabling tracking, branching, and reproducibility. Red Hat OpenShift AI delivers a scalable infrastructure, optimized container orchestration, and built-in AI/ML tools. Together, they transform AI pipelines, enabling faster iteration, better data quality, and full traceability—all while ensuring secure, enterprise-grade AI workloads. Why does this matter? Open-source advantage → Reduces vendor lock-ins and fosters community-driven innovation. Data Science efficiency → Supports version control, automation, and reproducibility. Scalable governance → Ensures transparency and compliance with detailed data lineage. With lakeFS and OpenShift AI, teams get the flexibility, control, and efficiency they need to drive real AI impact. Special thank you to [insert LI profile link / tag Sean Merrow- https://lnkd.in/dKEwRzbC] for helping this integration come to fruition!

127

5 Comments

Folks who run Kubernetes know that it is not the easiest software to manage, especially at scale. We built the DigitalOcean Kubernetes Service (DOKS) to bring simplicity to Kubernetes management, which has made a huge difference for developers across the globe. Today, I am excited to share with you some of the new enhancements we have made to DOKS to provide you an experience that effortlessly adapts to your growing needs -- we have doubled our max cluster capacity, added ability to seamlessly integrate with your VPC resources, introduced eBPF based routing to significantly improve bandwidth efficiency and network latency, and enhanced security, reliability, and scalability with managed Cilium. If you are interesting in learning more, check out our blog - https://lnkd.in/druhbjUF #Kubernetes #DigitalOcean #Cloud

36

The new best way to run Wasm on AKS is with SpinKube! Microsoft has sunset their older node pools in favor of this Cloud Native Computing Foundation (CNCF) project that Fermyon Technologies, Microsoft, SUSE, and Liquid Reply co-created. https://lnkd.in/gB97pnWF

36

1 Comment

"Azure Managed Redis overhauls in-memory caching with enterprise-level performance, flexible clustering, and AI-ready capabilities. It optimizes every CPU cycle, scales effortlessly, and integrates with existing tools, offering tailored solutions for modern, data-intensive cloud-native applications."

38

From Kubernetes diffs and ECS weighted traffic shifting to Terraform auto-approvals and GitOps enhancements, we packed Q1 with powerful CD platform upgrades. Highlights include: ✅ Namespace enforcement for Kubernetes & Helm. ✅ Blue-Green weighted traffic for ECS. ✅ Git-triggered pipelines in GitLab. ✅ Environment-type-based freeze windows. ✅ API-driven observability. 👉 Dive into what’s new: https://lnkd.in/dKGG3h95

24

Nice blog post by Nasuni's John Capello on disaggregated storage and why Nasuni leads the way 👉 https://lnkd.in/e8ge9QcU I'd be remiss if I did not also point out the benefits to #AI of a disaggregated storage architecture like #Nasuni's. As John points out in his post, Nasuni is a textbook example of a file-centric disaggregated architecture. The UniFS global file system lives as economical object storage in the cloud, while lightweight, stateless edge appliances supply whatever performance and protocol mix a site or workload needs. Capacity is separated from performance letting metadata be served from these purpose-built edges. Because the two layers scale independently, an enterprise can add petabytes of capacity for pennies per GB and/or deploy extra edge VMs for cloud or local AI-inference. Nasuni’s architecture continuously streams every new or modified file from each Edge Appliance into a single, cloud-resident UniFS globally distributed file namespace. Although this may seem simple, it removes one of the hardest obstacles that companies have in production AI ie. keeping models and pipelines fed with the latest data, no matter where in the world it was created. Nasuni’s edge-to-namespace replication turns the 'plumbing problem' of getting fresh, geographically dispersed data into an AI-ready repository into a solved service, letting companies focus on leveraging AI models with their data, not file logistics. #EnterpriseAI #DisaggregatedArchitecture #AIInfrastructure #AIPowered #GenAI #DataForAI #UnifiedFileSystem #CloudFileServices #EdgeData #GlobalNamespace #HybridCloud

22