Giant Conversations Episode #28
Episode Date: February 12 2025
Topic:
Hosted by:
Special Guest: Puja Abbassi
Contributors:
### Swarmalicious News
Zach Stone - TIL about this k8s native global load balancer project, currently CNCF sandbox: A cloud native Kubernetes Global Balancer
Advantages:
- Kubernetes Native Integration
- High Availability and Automatic Failover
- Flexible Load Balancing Strategies
New Applications:
- Multi-Cluster Deployments
- Integration with Liqo
- Enhanced Observability
Puja Abbassi - Our friends from Polar Signals built a nice tool to detect cross-zone traffic: eBPF Tool Identifies Cross-Zone Kubernetes Network Traffic
From The New Stack: Polar Signals couldn't figure out why it had so much expensive cross-zone traffic. So it turned to eBPF to deliver the answers, and cut its cloud bill in half.
Martin Proffitt - Why I Changed My Mind About Backstage? A Brutally Honest Opinion
What is his brutally honest opinion? It IS difficult to work with. But so was Kubernetes in the past. Better options back then would have been Docker, Mesosphere, etc. But NOW K8s is the defacto standard. Port for example may be easier now, but he believes Backstage will become the main choice in the future. Predominantly because of CNCF involvement.
Marcus Noble - Kubernetes History Inspector
This tool parses Kubernetes and Google Kubernetes Engine (GKE) logs, generating a comprehensive timeline of all events within a cluster.
Key Features:
- Comprehensive Event Timeline
- Information-Dense Interface
Xavier Avrillier - Cilium Releases v1.17.0: https://github.com/cilium/cilium/releases/tag/v1.17.0
The new Cilium version includes some pretty cool new features including:
- Quality of Service: Annotate your Pods for Guaranteed, Burstable or BestEffort egress network traffic priority
- FQDN Filtering for hostNetwork: Use CiliumClusterwideNetworkPolicy to configure Layer 7 filtering for DNS requests on nodes in the cluster
AI News
Timo Derstappen: - Let's different llms talk to each other to find consensus. llm-consortium
Key Features: Multi-Model Orchestration: The system coordinates responses from multiple LLMs simultaneously, allowing each model to contribute its unique strengths to the problem-solving process.
Iterative Refinement: Through multiple rounds of evaluation, the consortium refines responses to improve accuracy and coherence, ensuring that the final output is well-vetted.
Advanced Arbitration: A designated arbiter model synthesizes responses from various LLMs, evaluating and combining them to produce optimal results.
Blockchain Verification: Decisions and consensus are recorded on the Solana blockchain, providing transparency and verifiability of AI-generated outputs.
Puja Abbassi - DeepSeek claims its ‘reasoning’ model beats OpenAI’s o1 on certain benchmarks Chinese AI lab DeepSeek has released an open version of DeepSeek-R1, its so-called reasoning model, that it claims performs as well as OpenAI’s o1 on certain AI benchmarks.
Tim from Netlight mentioned is running most of his LLM workloads on K8s through: https://github.com/vllm-project/vllm It's a bit complex, but it supposedly what OpenAI also uses under the hood and very very powerful: DOCS
Timo Derstappen - Openai launched their first agent: https://www.youtube.com/live/CSE77wAdDLg
Simon Weald - Trolling SMS spammers with Ollama: https://evan.widloski.com/software/sms_llm/