Compute Service Selection

Huawei Cloud provides the Elastic Cloud Server (ECS) and Cloud Container Engine (CCE) services. The following table lists the ECS types provided by Huawei Cloud to meet the requirements of diversified computing scenarios.

For details about the preceding ECS types, see ECS Types.

**Table 1** ECS type
Architecture	ECS Type	Instance Family	Description	Scenario
x86	General computing-plus	c	A balance of compute, storage, and network performance, dedicated CPU, and stable performance	Applicable to most application scenarios
	General computing-plus	ac	Compared with C series, smaller network bandwidth is allocated to different CPUs and the same specifications, ensuring stable performance and lower costs.	Applicable to most application scenarios
	High-performance computing	h	Compared with the C series, higher CPU frequency delivers 20% higher compute performance.	HPC/Gaming/Scientific computing
	Memory-optimized	m	Compared with the C series, the memory-optimized ECSs with a CPU/memory ratio of 1:8 provide higher memory performance.	Memory-intensive and database/memory database
	Memory-optimized	am	Compared with the AC series, the memory-optimized ECSs with a CPU/memory ratio of 1:8 provide higher memory performance.
	Large-memory	e	Compared with the C series, the memory-optimized ECSs with a CPU/memory ratio of 1:20 provide higher memory performance.
	Disk-intensive	d	Compared with the C series, the large-capacity and low-cost SATA local disks are provided.	Big data/Cached database
	Ultra-high I/O	i	Compared with the C series, large-capacity NVMe local disks with high IOPS and low latency are provided.
	Ultra-high I/O	ir	Compared with the C series, small-capacity NVMe local disks with high IOPS and low latency are provided.
	General computing	s	Compared with the C series, General computing ECSs use a CPU-unbound scheduling scheme. When the host load is light, the same computing performance as the C series can be provided. The cost is lower. However, the stability of computing performance cannot be guaranteed. It is suitable for scenarios that can tolerate performance jitter.	General web/Development environment/Small database
	General computing-basic	t	Burstable performance instances with low costs. The burstable duration is determined by CPU credits.	Personal use/Maintenance terminal
	GPU-accelerated	g	T4 GPU for image acceleration	3D animation rendering, CAD, and more
		p	V100 GPU for computing acceleration	AI deep learning and scientific computing
		pi	T4 GPU for inference acceleration	Real-time inference + light-load training
	AI-accelerated	ai	Ascend 310 for computing or inference acceleration	Deep learning, scientific computing, and CAE
ARM	Kunpeng general computing-plus	kc	Compared with the C series, the Kunpeng processor is used, and the price is lower.	Adapted to most Arm application scenarios
	Kunpeng memory-optimized ECSs	km	Compared with the M series, the Kunpeng processor is used, which is more cost-effective.	Database/Memory database
	Kunpeng Ultra-high I/O ECSs	ki	Compared with the i series, the Kunpeng processor is used, and the price is lower.	Big data/Cached database
	Kunpeng AI Inference-accelerated ECSs	kai	Compared with the AI series, the Kunpeng processor is used, and the price is lower.	Deep learning, scientific computing, and CAE

The following describes the ECS selection principles:

Service applicability: Meet service requirements is the first principle for selection. In addition to CPU and memory, pay attention to bandwidth requirements. Generally, the larger the instance specifications of the same series, the higher the bandwidth allowed.
Cost-effectiveness: A cost-effective solution must be selected if service requirements can be met. For example, with the same specifications, the price of the S/AC series is lower than that of the C series. If there is no high performance requirement for O&M terminals, the T series is more cost-effective. For fluctuates services, you are advised to use multi-node cluster load sharing and AS. In this scenario, you are not advised to use high-specification instance nodes. Otherwise, performance will be wasted when the number of nodes is reduced to the minimum.
Reliability: Consider how to reduce the failure rate and prevent single points of failure when selecting resources. Therefore, you are advised to select the new series (with a larger number in the specifications) and deploy the resources in a balanced manner across two AZs. Resource selection optimization and cost reduction cannot compromise service reliability. The failure of a single node in a cluster network should not cause overload of other nodes.
Consistency: To ensure fast scale-out, fast recovery, and auto scaling based on images, hosts that carry the same type of services must have the same specifications. Do not use too many instance types or specifications in the same service system unless otherwise specified.
Resource satisfaction: Considering service development and scale-out requirements, you are advised to select mainstream models in mainstream AZs (for example, AZ 1 and AZ 7 in CN North-Beijing4 and AZ 1 and AZ 4 in CN East-Shanghai1) and avoid selecting old flavors.

BMSs are required in special scenarios such as AI. Generally, ECSs are used for general-purpose computing. The recommended models for typical scenarios are as follows:

**Table 2** ECS selection for typical scenarios
Location	Typical Application		Selection Suggestion
Access layer	Load balancing/Application proxy	Nginx	C/M series
Access layer	O&M terminal	Jump server	T series
Application layer	Common application	Web services	AC/AM series
Application layer	High-performance computing service	Transcoding	C/M series
Middleware	Self-built middleware	Self-built Redis/RocketMQ	C/M series
Data layer	Self-built databases	Self-built MySQL/Oracle	C/M series