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 –>
Highload
Highload on ThecoreGrid focuses on designing and operating systems that handle massive scale, traffic, and data under strict reliability requirements.
We explore architectures and patterns for horizontal scaling, load distribution, fault tolerance, and performance optimization in distributed environments. Topics include sharding, replication, caching strategies, queueing systems, backpressure handling, and latency reduction under peak load. We analyze real-world trade-offs between consistency, availability, and cost, along with failure scenarios and recovery strategies. Content is grounded in BigTech practices, including incident post-mortems and lessons from operating systems at global scale. You’ll find deep dives into infrastructure behavior, traffic management, autoscaling, and resilience engineering. Instead of simplified guides, the Highload tag delivers practical engineering insights for backend engineers, architects, platform teams, and SREs responsible for building and maintaining systems that must perform reliably under extreme demand.
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
MoE Expert Replication Without Excess Memory
How to optimize MoE expert replication: an analysis of CRAFT, load balancing, and throughput growth without overspending GPU memory.
Hybrid fronthaul planning in O-RAN networks
Hybrid fronthaul planning in O-RAN: how to reduce TCO and ensure capacity in CF-mMIMO through a combination of fiber, mmWave, and FSO.
Osprey event engine for real-time rule processing
Osprey event engine: how real-time event processing and rule evaluation work under high load, and what architectural trade-offs are hidden within the system
ParaQAOA Accelerates QAOA for Max-Cut
arXiv is the largest open preprint repository (since 1991, under the auspices of Cornell), where researchers quickly post working versions of papers; the materials are publicly accessible but do not undergo full peer review, so results should be considered preliminary and, where possible, checked against updated versions or peer‑reviewed journals.