Configuration
Kubernetes Base Environment Requirements
· ☕ 4 min read
1. Base Packages Kubernetes version ≥ 1.18 socat Required conntrack Required ebtables Optional, but recommended ipset Optional, but recommended ipvsadm Optional, but recommended 1 apt-get -y install socat conntrack ebtables ipset ipvsadm 2. Port Requirements 2.1 Base Node Services Service Protocol Action Start Port End Port Comment ssh TCP allow 22 22 Node remote management docker TCP allow 2375 2376 Docker remote communication etcd TCP allow 2379 2380 etcd cluster communication 2.

NVIDIA Environment Variable Configuration
· ☕ 2 min read
NVIDIA_VISIBLE_DEVICES Specifies the GPU devices visible to the program 1 CUDA_VISIBLE_DEVICES=0,1 Available values: 1,2, a comma-separated list of GPU UUIDs or indices all, all GPUs none, load the driver but with no GPU access void, do not load the driver NVIDIA_DRIVER_CAPABILITIES Controls which driver libraries/binaries will be installed inside the container

Ascend Environment Variable Configuration
· ☕ 4 min read
HCCL_IF_IP Configures the IP of the root communication NIC for HCCL initialization. Priority: environment variable HCCL_IF_IP > environment variable HCCL_SOCKET_IFNAME > NICs other than docker/lo (in ascending lexicographic order of NIC name) > docker NIC > lo NIC. 1 export HCCL_IF_IP=10.0.0.1 HCCL_IF_BASE_PORT Specifies the starting port number of the Host NIC.

Distributing Cluster Images with Dragonfly V2
· ☕ 10 min read
1. Introduction to Dragonfly The Dragonfly documentation at https://d7y.io/zh/docs/ already covers things in detail. Here is just a brief introduction to the main components of V2: Manager, which provides the UI, user management, cluster monitoring, task management, and other features Scheduler, which schedules traffic between Peers and provides preheating and other features Seed Peer, the back-to-source node used to download data from origin sites (Harbor, Docker.

Nydus Lazy-Loading Image Configuration and Practice
· ☕ 9 min read
Statistics show that most files inside a container are never used. Based on this characteristic, Nydus defines a custom filesystem in the Rafs format, enabling on-demand loading of image files to solve the slow startup and storage consumption caused by large images. In AI scenarios, whether for inference or training, images often start at several GB, or even tens of GB, so Nydus is very well suited.