CoreWeave plunges 20% after Q2 results, sparking debate on compute capacity vs capability.

We scaled #Keycloak to 2000 logins per second and 10000 token refreshes per seconds on #AWS and it scaled linearly in our #loadtest! It performed just fine with the latest optimizations in the latest 26.4 release. What load do you run with Keycloak? Do you host your own #IAM? Spoiler: Most of the CPU is burned for crypto operations (password hashing, signatures, TLS)

Today’s Q: What is the primary goal of implementing an Activity-Based Costing (ABC) system? A. To assign OVH to products based on the activities that consume resources. B. To calculate period costs more accurately than product costs. C. To reduce the total amount of OVH incurred by the company. D. To eliminate the need for any arbitrary cost allocation methods. ………………………….. #Daily_Dialogue

Wait time. What many people mean when they say "cut costs", they mean get things cheaper from vendors. Reduce operating expenses. Put off or reduce capital purchases by across the board 20% budget cuts. Ignored and not considered is wait time. Technology's biggest value is reducing it. Performance problems are often identified by "it feels like it's going slower" or actually pulling out a stopwatch and timing an operation, or using a time function. As I sit here waiting for a virtual machine to copy, a document to upload, a little application to finish compiling, and authenticate for the fifth time this hour, I reflect on how much time is spent waiting. If you have a team of 8 developers who wait, say 3 minutes per hour, that's 16 hours of wait time per week. $1,600 a week. $83,200 of non-productive time per year. But we want to quibble over whether some widget boops and beeps for a bit more than we want to pay. In fact wait time is really THE consideration. And ironically it's not the speed of the beeps and the boops, it's the operating policies that govern the operation of the beeps and the boops that limit how many beep and how many boops it can realistically produce. If you want to be "efficient" and reduce your operating expenses, start with your operational policies. Those are the "software" of your business processes. In case you are wondering the VM took 25 minutes to clone, the application took 8 minutes to compile, the document got stuck uploading, and I just had to authenticate again for the 7th time.

🇧🇷 **ram** Category : Business and Finance | Pays : BR ## RAM: The Silent Engine Driving Brazil's Digital Transformation In today's fast-paced Brazilian business landscape, digital efficiency is paramount. While often overlooked, Random Access Memory (RAM) is a critical component powering everything from daily operations to cutting-edge innovations. It’s more than just a specification; it’s a foundational element of your business’s speed and agility. Think of RAM as your business's short-term memory – the faster and larger it is, the more tasks your systems can handle concurrently without slowing down. For Brazilian enterprises, this translates directly to productivity gains. Whether it’s accelerating complex financial models, enhancing customer experience in e-commerce, or running sophisticated data analytics for market insights, sufficient RAM ensures applications run smoothly, reducing frustrating lags and boosting employee output. Investing strategically in RAM is no longer just an IT concern; it's a smart business decision. As cloud adoption soars in Brazil and AI-driven solutions become mainstream, the demand for robust memory will only intensify. Adequate RAM future-proofs your infrastructure, supports scalability for growth, and allows your teams to leverage advanced technologies without performance bottlenecks. It’s an investment that pays dividends in efficiency, innovation, and competitive edge. Don't let insufficient RAM be a hidden drag on your business's potential. Evaluate your current infrastructure and consider how optimized memory can unlock greater speed, efficiency, and innovation. What impact has RAM had on your business operations? Share your thoughts below! #RAM #BusinessTechnology #DigitalTransformation #CloudComputing #DataAnalytics #ITInfrastructure #BrazilBusiness #FintechBR #TechTrends #Productivity #Investment #Business and Finance #BR #trends Sources: - https://lnkd.in/eDQJJnHn

There are rumours that CoreWeave is running a Ponzi scheme. This has come about after they released a report into its Q2 2025 earnings. Is this true? No. In short, the slide reveals efficient cash management for growth, not fraud. If new details emerge (e.g., delivery failures), that could change things—but as of October 2, 2025, this is overblown hype. What the Slide Actually Shows The relevant slide (Slide 12: "Contract Mechanics: Understanding CoreWeave’s Business Model") outlines a typical contract lifecycle timeline: Month 1 (Contract Signing): Customers make an initial prepayment (cash inflow to CoreWeave). Month 3: CoreWeave submits a purchase order for systems and makes cash payments for GPUs and other infrastructure upon delivery (cash outflow). Months 3–5: ~3-month installation period for the infrastructure. Months 6–30+: AI cloud services go live, with revenue recognised rateably (evenly) each month via ongoing customer payments (cash inflows).

“Just lift and shift.” Fast, simple, done. Right? Lift & shift buys speed, but costs compounding margin erosion. Phased refactor feels slower, but lets you tune infra to economics as you scale. Greenfield sounds scary, but sometimes it’s the only way to avoid a cost ceiling. CTOs don’t talk enough about the unit cost per user when making these calls. That’s the only metric that cuts through the noise. Speed now vs. margin forever - that’s the choice. #CloudMigration #Scalability #CloudStrategy

What we’re seeing in the market is not unification, but its simulation. Vendors are stitching layers together just enough to market the story of simplification, while structurally increasing dependence on their ecosystem. The pattern repeats across the stack: https://lnkd.in/gG_tg8Dh

🌟 New Blog Just Published! 🌟 📌 pgEdge Drops Proprietary Model, Embraces Open Source 🚀 ✍️ Author: Hiren Dave 📖 pgEdge was founded at a time when the broader database market seemed to have settled into a comfortable equilibrium of proprietary licensing and closed-source extensions. While most vendors...... 🕒 Published: 2025-09-28 📂 Category: Tech 🔗 Read more: https://lnkd.in/d27tgYZ6 🚀✨ #pgedge #opensource #postgres

Just wrapped up a distributed inventory system kata based on "Designing Data-Intensive Applications" Chapter 8. Built a reservation system that prevents overselling during flash sales, even with network delays and concurrent requests. The key was implementing atomic reservations with cleanup on failure. Used a two-phase approach: TryReserve creates a temporary hold, then ConfirmReservation finalizes it. NetworkLagDB simulates 50-200ms delays to test real-world conditions. Interesting sync issue: initially used defer for reservation cleanup, but that released even successful orders. Switched to explicit ReleaseReservation calls only on confirmation failures. Also had to be careful with mutex scopes - Query() locks/unlocks separately from TryReserve's lock. The reservation pattern handles the fundamental distributed systems problem: multiple services seeing stale inventory data due to network partitions. Each reservation gets a unique atomic ID, and the system maintains consistency without relying on global clocks. Code: https://lnkd.in/d4Vxwdt9 #Go #DistributedSystems #Concurrency #SystemDesign

Reliability in Web3 infrastructure is non-negotiable. Which RPC Proxy delivers? We ran the load tests on Kubernetes to find out. For enterprises and infrastructure providers running critical Web3 operations, the choice between tools like aetherlay and dshackle defines your uptime and error rates. Our latest post dives deep, complete with a head-to-head benchmark using k6 on a minikube setup. We assigned equal resource budgets to both services to truly test their out-of-the-box performance. 🖊️ Here's the critical takeaway: Under Load: aetherlay maintained near-perfect stability (0% errors at 20 VUs), handling the load with increased but predictable latency. The Trade-off: dshackle delivered faster successful requests, but it struggled significantly with stability, requiring manual intervention to recover after crashing. 🤔 Our Conclusion: If you value reliability and a Kubernetes-native, low-maintenance approach for your service redundancy, you need to read these results. 📰 Full Report & Benchmark Data: https://lnkd.in/evag-FMX