HCIP-Cloud Computing cheat sheet

Computing: Huawei Cloud Compute Services and Architecture

• ECS instance families: general-purpose (s/c), memory-optimized (m), compute-optimized (c), GPU-accelerated (g/p), and high-performance (h); the letter prefix identifies the family and the trailing number the generation.

• Flavor selection rule: match vCPU-to-memory ratio to workload - general-purpose is roughly 1:2 or 1:4, memory-optimized 1:8, compute-optimized around 1:1 to 1:2.

• ECS billing modes: pay-per-use (hourly/second-level), yearly/monthly subscription (discounted, suited to stable long-running workloads), and spot/preemptible instances for fault-tolerant batch jobs.

• An ECS instance is bound to an AZ at creation and cannot be migrated across AZs without a rebuild or image-based redeployment; the VPC subnet chosen at creation fixes the AZ.

• EVS disk types: high I/O (SSD-backed, low latency), ultra-high I/O (NVMe-class), general-purpose SSD, and standard; disk performance scales with capacity, so small disks cap IOPS.

• AS (Auto Scaling) components: scaling configuration (template of flavor, image, disk, key pair), scaling group (min/desired/max instances, VPC subnet, load balancer binding), and scaling policy (scheduled, alarm-based, or periodic).

• Scaling policy types: alarm (CES metric threshold), scheduled (cron-like time window), and lifecycle hooks that run a script or notification before instance add/remove.

• BMS (Bare Metal Server) is chosen when the workload needs no hypervisor overhead, non-virtualized hardware access, or high-performance databases and HPC; it still integrates with VPC, EVS, and ELB.

• DeH (Dedicated Host) provides physically isolated hosts for licensing and compliance; ECS instances on a DeH share the host's physical resources and support host-level affinity/anti-affinity.