Network slicing with trust constraints without latency loss
Online network slicing with trust constraints: how the Path–Link model reduces latency and accelerates VNF placement in multi-domain infrastructure.
Architecture and Infra
Architecture and Infra on ThecoreGrid covers the foundations of designing and operating scalable, reliable systems at BigTech level. This category brings together system design and infrastructure practices: distributed architectures, highload patterns, cloud-native platforms, and core layers such as compute, networking, and storage. We focus on real engineering decisions — how to balance reliability, performance, cost, and long-term system evolution. Topics include Infrastructure as Code, Kubernetes, multi-region deployments, traffic management, and platform design. Content is grounded in production experience: incident post-mortems, large-scale migrations, and lessons from operating infrastructure under heavy load. Instead of abstract theory, you get practical trade-offs, proven patterns, and insights drawn from real-world systems. Architecture & Infra is built for architects, backend and platform engineers, DevOps teams, and SREs responsible for complex distributed systems and mission-critical infrastructure.
Reverse Address Translation in multi-GPU systems
How Reverse Address Translation affects latency in multi-GPU systems and why TLB misses hinder All-to-All operations in ML workloads.
Slice spraying in GPU clusters without blocking
Slice spraying in GPU clusters: how TENT reduces latency and increases throughput in LLM serving through dynamic data movement –>
Distributed sequence generation without bottlenecks
Distributed sequence generation replaces database sequences at scale. It removes central bottlenecks while keeping compatibility with existing systems. The problem does not manifest immediately — until the organization attempts to transition from a relational database to a cloud-native storage solution. In this case, over a hundred services relied on database sequences for generating primary keys. … Read more
Multi-path GPU Traffic Balancing Without Bottlenecks
Multi-path GPU balancing eliminates network bottlenecks in clusters. An analysis of NIMBLE and its impact on throughput and latency. –>
GitOps Policy for Kubernetes through Kyverno and Argo CD
GitOps policy for Kubernetes becomes manageable when enforcement is built into the delivery pipeline. The combination of Kyverno and Argo CD bridges this gap at the admission level.
SKID Identifier System for Distributed Systems
SKID identifiers: how to combine sortability, security, and zero-lookup verification in distributed systems without dual keys. –>
ThecoreGrid Radar: LLM infrastructure moves into disaggregation, networks become programmable, and distributed systems formally verifiable
LLM Infrastructure, Disaggregation, Distributed Systems, GPU Clusters, Network Anomalies, Serverless, AI Agents
LLM evaluation at scale on Apache Spark
LLM evaluation at scale on Apache Spark: how the distributed architecture, caching, and statistical validation of models are structured.
Scaling Uber: Systems, Teams, and AI Engineering
Ex‑Uber CTO Thuan Pham on scaling systems, microservices, platform teams, and using AI to transform software engineering.