news
Red Hat Leftovers, Lots of Slop Promotion (Still)
-
Red Hat ☛ Using NetworkManager to permanently set an interface administratively down
This article demonstrates the legacy way and a new way to permanently set a network interface administratively down with NetworkManager and
nmclicommands.Since the very first NetworkManager announcement in 2004 (figure 1), a key property of NetworkManager has been the ability to set all managed interfaces that sense carrier to
up. This makes sense with desktops that need to juggle Wi-Fi and wired connections. But sometimes, especially with servers and appliances, we want interfaces to always staydownacross bootstraps, even if they sense carrier. Maybe an interface is unused and we want to make sure nothing bad happens if somebody mistakenly connects a cable to it. Or maybe it's part of a manual failover design and needs to bedownuntil somebody sets it toup. -
Red Hat ☛ MPI-powered gradient synchronization in PyTorch distributed training
The theoretical capability of a 1,000-GPU cluster is staggering, but in practice, the hardware often spends more time "talking" than "thinking." In distributed training, the gradient synchronization phase is the bottleneck. If one node finishes its backward pass 10 minutes late, the entire cluster must wait, which causes prolonged idle compute time.
To turn a collection of isolated GPUs into a single cohesive learner, we rely on Message Passing Interface (MPI)-style collective operations. Specifically,
All-Reduce—which sums gradients across all nodes and redistributes the result—must be executed with precision to prevent the network from becoming a graveyard for throughput.In this article, we move from theory to implementation. We will look deeper into gradient synchronization, explore the core MPI primitives that power it, and walk through a production-ready Pytorch implementation using the torch.distributed backend to bridge the gap between code and massive-scale hardware.
-
Red Hat Official ☛ Securing the enterprise software fabric: A blueprint for open source
Open source effectively serves as the foundation for all of modern technology, not just enterprise IT, and this is about much more than just Linux. Application servers, databases, network routing, developer environments, and all of the other invisible components that make up our technological fabric are fueled by open source projects in some way, shape, or form.
-
Red Hat Official ☛ Scaling automated infrastructure compliance in telecommunications using Red Hat Ansible Automation Platform
To support a connected future, Telstra needed to move beyond manual intervention and embrace a strategy where the network could manage, protect, and optimize itself. Telstra modernized its approach by transitioning to Red Hat Ansible Automation Platform, shifting from monolithic playbooks to a modular, reusable automation architecture.
-
Red Hat Official ☛ Protecting enterprise AI: How to manage API keys in Models-as-a-Service (MaaS) [Ed: Slop promotion by Red Hat]
What happens next is predictable. A developer's personal token gets copied into a secret. Or a service account gets created with more access than anyone intended, shared across pipelines, and forgotten. The traffic flows. Nobody knows whose budget it counts against. When that developer moves teams 6 months later, the credential lives on in places nobody knows about.
-
Red Hat ☛ llama.cpp vs. vLLM: Choosing the right local LLM inference engine [Ed: IBM Red Hat promoting slop and GAFAM]
Everyone wants to run local large language models (LLMs) right now, and for good reason. Running your own GPT-style models means no API bills creeping up month over month, no rate limits from a model vendor, and full data privacy by default. Whether you're building retrieval-augmented generation (RAG) pipelines, spinning up AI agents, or using an AI code assistant, two inference engines keep coming up in most conversations: llama.cpp and vLLM.