Weekly Cloud Info #W48 - 2024

Author

Walid LARABI
November 29, 2024

Hi!

Welcome to this week's cloud roundup! Highlights include AWS PreInvent updates like PrivateLink cross-region support, FSx for Lustre’s 12x GPU performance boost, Cognito's passwordless login, and VMware workload simplification. In AI, Alibaba unveiled the QwQ-32B model, and Microsoft introduced Graph RAG for better question answering. Kubernetes addressed memory challenges and launched Gateway API v1.2.

There is also an exciting Kubernetes certifications promo at the end of the email!

Have a great read. ☕

📰 Top picks of the week

AWS PrivateLink Launches Cross-Region Connectivity for Enhanced Security and Simplicity

AWS PrivateLink now allows cross-region connectivity for Interface VPC endpoints, enabling connections to VPC endpoint services in different regions. Providers can offer their services without needing extra infrastructure, while consumers can connect securely without public internet exposure. This feature is available in multiple regions, including the US, Europe, and Asia. Pricing details are available on the AWS PrivateLink pricing page.

AWS Boosts FSx for Lustre Performance: 12x Higher Throughput for GPU Instances

Amazon FSx for Lustre now supports Elastic Fabric Adapter (EFA) and NVIDIA GPUDirect Storage (GDS), boosting storage performance for GPU instances in the cloud. This enhancement allows for up to 12x higher throughput per client instance, reaching 1200 Gbps. EFA optimizes network throughput while GDS enables direct data transfer to GPU memory, reducing workload costs and speeding up machine learning training. The new features are available at no extra cost on Persistent-2 file systems across AWS Regions.

AWS CloudFront Enhances Performance with Free Origin Modifications via CloudFront Functions

Amazon CloudFront now allows origin modifications through CloudFront Functions, enabling custom routing based on request conditions. Users can overwrite origin properties and forward requests to various HTTP endpoints, ensuring lower latency by directing traffic to the nearest AWS Region. This feature enhances performance and decreases costs compared to using AWS Lambda@Edge. Origin modifications come with no extra charges and support existing origin capabilities.

AWS Enhances Amazon Cognito with Customization, Passwordless Login, and New Pricing Tiers

Amazon Cognito has launched major updates to enhance app authentication, including a new developer console for easier setup. The Managed Login feature offers customizable sign-in options, and support for passwordless login is now available. New pricing tiers (Lite, Essentials, and Plus) cater to various use cases. These improvements aim to provide better security and user experience for developers and their applications.

AWS Launches Amazon Elastic VMware Service for Simplified Workload Migration

AWS and VMware are advancing their collaboration to support VMware workload migration to AWS. The newly launched Amazon Elastic VMware Service (Amazon EVS) allows customers to run VMware Cloud Foundation on AWS with ease, ensuring compatibility and simplifying deployments. This service enables quick setup and migration without major changes or retraining. AWS aims to enhance the cloud experience for VMware users while facilitating modernization.

Alibaba Unveils QwQ-32B-Preview AI Model with 32.5 Billion Parameters

Alibaba has launched a new AI model called QwQ-32B-Preview, featuring 32.5 billion parameters and the ability to process prompts up to 32,000 words. It outperforms some OpenAI models on specific benchmarks like AIME and MATH, showcasing strong logic and math skills. However, it may struggle with common sense reasoning and language switches. The model is available for download under a permissive license but contains some restricted components.

Microsoft Unveils Graph RAG for Enhanced Question Answering Over Large Texts

Microsoft proposes a new method called Graph RAG that improves question answering over large text corpora using retrieval-augmented generation. This approach builds an entity knowledge graph and pregenerates community summaries to enhance response quality for broad questions. It demonstrates significant improvements in answer comprehensiveness and diversity compared to traditional RAG methods. An open-source implementation for both global and local approaches will be available soon.

Microsoft Faces 11-Hour Global Outage Affecting Exchange Online and Teams

Microsoft experienced a global outage affecting Exchange Online and Teams Calendar for over 11 hours, requiring manual server restarts. The company acknowledged issues with a recent change that caused the outage and struggled to restore full service. Users expressed frustration over untested changes impacting live services. Recovery efforts were noted but took longer than expected.

CNCF: Gateway API v1.2 Launches with New Features and Breaking Changes

The Kubernetes SIG Network has released Gateway API v1.2, which includes new features like WebSockets, timeouts, and retries, while retiring old versions of GRPCRoute and ReferenceGrant. This update also introduces breaking changes, such as modifications to the format of supported features. Four features have graduated to standard availability, enhancing the API's reliability. Users are advised to check compatibility before upgrading.

Kubernetes tackling kube-apiserver Memory Issues with Proposed LIST Request Optimization

The kube-apiserver faces critical memory explosion issues due to unpredictable memory consumption from LIST requests, especially in large clusters. This can lead to server slowdowns, resource pressure, and workload disruptions. A proposal aims to reduce memory usage considerably, protecting the kube-apiserver from out-of-memory (OOM) attacks and lowering the load on etcd by optimizing LIST requests. Goals include reducing temporary memory consumption.

❤️ You might also like

  • Amazon develops video AI model, The Information reports LINK

  • xAI could soon have its own chat app LINK

  • Former Android leaders are building an ‘operating system for AI agents’ LINK

  • Voice cloning, real-time transcription, translation, and TTS with support for voice separation, YouTube downloading, and podcast creation using celebrity voices on Windows 10/11 LINK

  • Artists leak OpenAI's Sora video model LINK

  • Anthropic says Claude AI can match your unique writing style. LINK

🎁 This week hidden gem

60% OFF promo from Linux Foundation (eg: 598$ for 5 K8S Certification) LINK

Send “GIFT” to [email protected] to receive all of this month's hidden gems within next 2 hours.

🏁 Enjoy this newsletter?

Forward it to a friend, and let them know they can subscribe here.