Zach S.

Bare Metal Kubernetes: Is the performance advantage disappearing? For years, running Kubernetes on bare metal was the go-to for performance-obsessed teams. No virtualization layers. No shared resources. Just raw speed. But the landscape is changing. Today’s cloud-native tooling, container runtimes, and orchestration improvements are closing the gap. Virtualized environments are now offering performance that's nearly indistinguishable—while still giving you scalability, flexibility, and manageability. CTOs and engineering leaders: It may be time to rethink that bare-metal loyalty. Unless you're running ultra-low latency workloads (think: trading platforms or real-time AI inference), the performance trade-off may no longer be worth the operational complexity. 📉 The marginal gains are shrinking. 📈 The cost of complexity is growing. What’s the real ROI of squeezing out the last 5% of performance—if it costs you 50% more in engineering overhead? 👉 Full breakdown here: https://lnkd.in/eYPACz5Q #Kubernetes #CloudNative #DevOps #CTOInsights #InfrastructureEngineering #PlatformEngineering #CloudComputing #PerformanceEngineering #SRE #BareMetal #TechStrategy #DigitalTransformation

6

6 Comments

Three projects exploring the application of QUIC to video broadcasting. One attempts to replace SRT and WHIP with live DASH-IF Ingest QUIC streams (quicfeed.net), another publishes and plays live CMAF via MoQ relays, and another is a QUIC-based system (packager/server/player) that is focused on exploring server-side ABR and ad-insertion as well as client synchronization via a wall clock. DM for demos or a PDF with details on my ABR implementation.

36

There is a cool new capability in the latest MCP spec called "tasks". I think it has real promise, but it's still broken in that it relies on polling. An MCP client has to connect and get status. https://lnkd.in/gERGDMaR This is one of the kinds of fundamental differences between the Developer interaction model with agents and the Operator interaction model with agents. Operators know that polling doesn't scale. Sure, it's great on your laptop! Try moving it to a shared endpoint and see what happens. We need MCP to move beyond these older school patterns into more modern cloud-native patterns. Loose-coupling, circuit-break, fan-out, etc. etc. There are already some moves in this direction: https://lnkd.in/gYv8Caif

6

1 Comment

AWS introduced EKS Capabilities, a new managed approach to run Kubernetes-native tooling without installing or operating it yourself (saves a lot of time). You can now enable Argo CD for GitOps, ACK to manage AWS resources from Kubernetes, or Kube Resource Orchestrator (KRO) for advanced workload orchestration. AWS hosts and maintains these components in isolated managed accounts, including scaling, patching, and upgrades. This reduces operational load for platform teams. No more managing controllers, worrying about upgrades, or debugging custom installs. It also standardizes the Kubernetes experience across clusters and environments. For anyone building a cloud platform on AWS, this is a clean way to simplify the stack and accelerate delivery while keeping a Kubernetes-first workflow. #AWS #AWSreinvent #EKS #GitOps https://lnkd.in/e-B52izx

9

Don't need an entire GPU for your workflow? AWS announces general availability of Amazon EC2 G6f instances, with partitions as small as 1/8 of an NVIDIA L4 Tensor Core GPU. https://lnkd.in/gsf47mqh

41

3 Comments

Over the past eight years I’ve worked across the open-source networking world—from building residential routers at Comcast, to Amazon Walkout Technology team, and most recently the Telecom Infra Project OpenWiFi initiative. One lesson has paid dividends in every role: really understanding how a packet traverses the Linux networking stack. If you want to build the same picture in your head, start with these two gold-standard resources: 1️⃣ “How packets travel through Linux” – a one-hour deep dive by Jiri Benc 🎥 https://lnkd.in/e4fzxiPb 2️⃣ linux-network-performance-parameters – Leandro Moreira GitHub guide that links every critical /proc, sysctl, and tc knob to the problems it solves 📚 https://lnkd.in/epKnTgCh Bookmark them—you’ll never look at packet flow the same way again.

109

7 Comments

Latency, efficiency, simplicity, and capabilities. These are the top 4 reasons users switch to ScyllaDB from Cassandra. If you're tired of unpredictable and sluggish performance, slow release cycles, and lack of critical features, see how companies like Discord, Fanatics, and Comcast found success after migrating to #ScyllaDB > https://lnkd.in/esKDRANj

82

Great to see the MCP spec and community evolving with a focus on production scaleability and laying the groundwork for collaboration that can actually scale. Good to see server identity (like .well-known) and async ops called out as priorities for the next release. https://lnkd.in/g9kQjmy4 #mcp #genai #sigs

51

2 Comments

A few days ago, I got an unexpected email from AWS Budgets with the subject: “Budget threshold exceeded – Learning AWS” I froze for a second. “Wait… what budget? I don’t remember setting one.” Immediately, my mind jumped to worst-case scenarios: 💣 “Did someone hack into my account?” 💸 “Are they mining crypto with the biggest EC2 instance money can buy?” ☠️ “Am I about to drain all my AWS credits in one night?” So I rushed to the console… and this is what I found (screenshot below 👇): 📉 A $5/month budget named Learning AWS, created back in 2019. It had never been exceeded—until now. This month, I had spent… $9.22. Once the panic settled, I had to laugh. In over 5 years, this was the first time I went over that tiny budget. But here’s the real insight: You can go a long way with the AWS Free Tier, credits, and a bit of curiosity. This alert was a reminder of just how much I’ve been able to explore, prototype, and learn—without ever spending more than a coffee or two per month. That $5 budget was never meant to be a limit… just a little safety net from a younger, cautious version of me. I’ll keep it there as a badge of honor now. A silent witness to hundreds of Lambda tests, DynamoDB experiments, IAM mistakes, and CloudFormation misadventures. If you’re just starting your AWS journey, don’t be afraid to dive in. Set a small budget. Get alerts. Break things. Learn fast. AWS is an incredible playground when used with care—and alerts like this one are great training wheels. 🛠️💡🚀 Can you guess which AWS service finally pushed me over the $5 budget this month? (Hint: it wasn’t EC2 👀) Drop your guesses or share your own “AWS budget alert” stories in the comments—I’d love to hear them! #AWS #LearnByDoing #CloudComputing #FreeTier #AWSCommunityBuilder

14

3 Comments

When designing scalable and resilient Kubernetes workloads, Pod Affinity and Pod Anti-Affinity play a critical role in how your pods are scheduled across nodes. Understanding them properly can significantly improve both performance and availability. 🔹 What is Pod Affinity? (Attract Pods Together) Pod Affinity ensures that a pod is scheduled close to other pods based on labels and topology. ✅ Key Features: 📌 Co-location of Pods: Places related services (e.g., app + cache) on the same node/zone 🏷️ Label-Based Matching: Uses labelSelector to identify target pods ⚡ Improved Performance: Reduces network latency for tightly coupled apps 🎯 Topology Awareness: Supports node, zone, and region placement ⚖️ Hard & Soft Rules: requiredDuringSchedulingIgnoredDuringExecution (strict) preferredDuringSchedulingIgnoredDuringExecution (flexible) 🔄 Dynamic Behavior: Rules apply only during scheduling, not enforced after 📌 Use Case: Microservices architecture where backend and cache (Redis) need fast communication. 🔹 What is Pod Anti-Affinity? (Keep Pods Apart) Pod Anti-Affinity ensures that pods are distributed across nodes/zones to avoid failures. ✅ Key Features: 🚫 Separation of Pods: Prevents similar pods from running on the same node 🛡️ High Availability (HA): Ensures resilience during node failures ⚠️ Prevents SPOF (Single Point of Failure) 🌍 Multi-Zone Distribution: Ideal for production-grade clusters ⚖️ Hard & Soft Rules Support 📉 Improves Fault Tolerance 📌 Use Case: Deploying multiple replicas of a web app across different nodes/zones. ⚖️ Affinity vs Anti-Affinity (Quick Comparison) Pod Affinity 👉 Performance Optimization Pod Anti-Affinity 👉 High Availability & Reliability Affinity 👉 Brings Pods Together Anti-Affinity 👉 Spreads Pods Apart 🛠️ Troubleshooting Common Issues 🔍 1. Pod Stuck in Pending Cause: No nodes match affinity rules Fix: Relax rules (use preferred), scale cluster, verify labels 🔍 2. Incorrect Label Selector Cause: Label mismatch Fix: Validate using kubectl get pods --show-labels 🔍 3. Topology Key Issues Cause: Invalid or missing topologyKey

25

I've become a huge fan of Command Line (CLI) Agents like Claude Code and Codex. They bring me back to the early days of my career when I first started using the terminal (via Putty) to SSH into servers, learn vim/grep/sed commands to edit, search files. Except, now the LLMs take care of running them and I simply observe and intervene once in a while. After movies like The Matrix and The Italian Job released, working on a terminal window with black and green lettering and results loading were seen to be "very cool"! At one of my workplaces, I had colleagues walk by my desk and be gobsmacked at what I was "hacking" together - when all I was doing was 'ls -ltr' to list directories 😆. What I realized then and what comes back again now is that for a lot of folks, the terminal is actually scary and too hard to even bother with. I recently read an article that highlights how the concept of folders, organizing folders in various directories and such is not known to many - and not taught in schools either. I don't know how true this is but I feel like one misses out on the various ways in which these CLI agents can be leveraged - not just for coding but also for more "office" tasks like creating and manipulating excel sheets and powerpoints. The article focusses on certain examples but I was curious if this is true across the world or if others have observed this behaviour?

30

1 Comment

🌐 AWS just rewrote how the cloud connects to everything. AWS announces two game-changing connectivity offerings: AWS Interconnect - last mile and AWS Interconnect - multicloud—eliminating weeks of complexity for hybrid and multicloud architectures. AWS Interconnect - last mile (Gated Preview) Connect branch offices, data centers, and remote locations to AWS in minutes, not weeks. ✅ Enter location → pick bandwidth → select region → instant connection ⚡ 1 Gbps to 100 Gbps on-demand scaling, zero downtime 🔒 MACsec encryption enabled by default 📍 Available now for US customers (Lumen partnership) AWS Interconnect - multicloud (Preview) Connect AWS to Google Cloud today, Azure later in 2026—purpose-built for multicloud strategies. 🔄 Private, secure, high-speed connections between AWS VPCs and other CSPs ⏱️ Deploy in days instead of weeks or months 🎯 Works with AWS Transit Gateway, Cloud WAN, and VPC 🏗️ Available in 5 AWS Regions Why this matters: For enterprises with hybrid infrastructure or multicloud strategies, this eliminates the DIY networking complexity that previously required months of planning and implementation. Now it's just a few clicks. Request access to last mile: https://lnkd.in/gN6VEqbH 🚀 Day 1 of re:Invent 2025 coverage — daily breakthroughs across compute, storage, database, networking, application development, analytics, and AI/ML. More coming! #AWS #Networking #HybridCloud #Multicloud #DirectConnect #CloudInfrastructure #reInvent

20

This is a huge deal for AI in enterprises. Architecting a global KV cache with Pure Storage eliminates redundant prefill, overcomes HBM limits, and delivers faster, more cost-efficient enterprise LLM inference at scale. See how you can keep your GPUs busy with KV Cache.

1

Compute infrastructure has a hidden bottleneck. Not CPU. Not memory. It's energy. Most systems still run as if power is infinite. Seamium introduces a governance layer that keeps workloads aligned with real energy conditions. See how it works: seamium.com #seamium

1

Managing your CDN with Infrastructure as Code? The Fastly Terraform Provider makes it easy to manage Fastly services directly from your Terraform workflows—so you can provision infrastructure, automate deployments, and keep your edge configuration version-controlled. One thing I appreciate: the provider is actively maintained with a high level of care by the Fastly team. Reliability and performance matter at every layer, and that attention to detail shows up even in the tooling developers rely on every day. If you’re integrating edge infrastructure into your DevOps pipelines, it’s worth checking out. Learn more and explore the docs: https://lnkd.in/eGFGfry3 #Terraform #DevOps #EdgeComputing #InfrastructureAsCode

5

𝗝𝘂𝘀𝘁 𝗺𝗼𝗱𝗲𝗹𝗲𝗱 𝘁𝗵𝗲 𝟯-𝘆𝗲𝗮𝗿 𝗧𝗖𝗢 𝗳𝗼𝗿 𝗮 𝟭𝟬𝟬-𝗰𝗵𝗮𝗻𝗻𝗲𝗹 𝗼𝗽𝗲𝗿𝗮𝘁𝗼𝗿 migrating from a legacy appliance to cloud vs on-prem. The gap is significant. StreamVX framed it perfectly: 𝗳𝗼𝗿 𝘁𝗵𝗮𝘁 𝗰𝗼𝘀𝘁 𝘆𝗼𝘂 𝗰𝗼𝘂𝗹𝗱 𝗿𝗲𝗽𝗹𝗮𝗰𝗲 𝘆𝗼𝘂𝗿 𝗲𝗻𝘁𝗶𝗿𝗲 𝗵𝗲𝗮𝗱𝗲𝗻𝗱 𝗮𝗻𝗱 𝗮𝗱𝗱 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆. If you have a similar platform in your headend, this is worth reading.

6

🌟 I would like to share some insights from the CDN Alliance's latest whitepaper on Low Latency Streaming! The industry has been grappling with standardizing low latency streaming definitions, and this comprehensive framework is a game-changer. Here's what you need to know: 🎯 The whitepaper introduces four clear categories of latency: 🔹 Real-Time Streaming (0-0.5s) 🔹 Sub-Second Streaming (0.5-1s) 🔹 Ultra-Low Latency (1-3s) 🔹 Low Latency (3-12s) 🔑 Key technologies like HESP, WebRTC, WebRTS, WebSockets  and Media over QUIC (MoQ) are transforming how we deliver content, each serving specific use cases from live sports to iGaming. 💡 The future of streaming is about balance - finding the sweet spot between latency and quality while ensuring security through DRM support. ☁️ Traditional CDNs work well for LL and ULL, while RTS and SSS often require specialty CDNs or cloud-based solutions This standardization effort by @CDN Alliance marks a significant step forward for the streaming industry. Whether you're in broadcasting, gaming, or live events, understanding these standards is crucial for delivering optimal viewing experiences. Some questions came up to me while reading the doc: 🔸 How will the emergence of Media over QUIC (MoQ) impact the current landscape of low latency streaming technologies once it becomes production-ready? 🔸 What specific technical challenges need to be overcome to achieve both low latency and high bitrates at scale? 🔸 How might the standardization of low latency terminology and definitions influence the development of future streaming technologies and industry adoption? 🏁 What are your thoughts on these new standards? Let's discuss in the comments! 👇 #Streaming #Technology #CDN #DigitalMedia #Innovation #TechStandards #Broadcasting https://lnkd.in/dKdiz8fx

2