Updated on 2023-12-22 GMT+08:00

ECS Types

The cloud platform provides the following ECS types for different application scenarios:

x86 and Kunpeng Architectures

ECS supports the following architectures:

  • x86 architecture

    The x86 architecture uses the complex Instruction Set Computer (CISC). CISC has a large collection of complex instructions that range from simple to very complex and specialized in the assembly language level, which takes a long time to execute the instructions.

  • Kunpeng architecture

    The Kunpeng architecture uses the reduced Instruction Set Computer (RISC). RISC a microprocessor architecture with a simple collection and highly customized set of instructions. It is built to minimize the instruction execution time by optimizing and limiting the number of instructions.

    The Kunpeng architecture delivers more balanced performance/power consumption than the x86 architecture.

Table 1 Comparison between the x86 and Kunpeng architectures

Dimension

x86 Architecture

Kunpeng Architecture

Advantage

Good ecosystem, support for almost all general software

Huawei-developed processors, cost-effective

Scenario

Platform-dependent scenarios using Windows software and x86-compatible commercial software

  • Platform-independent scenarios such as e-commerce, big data, and scientific computing
  • Arm native scenarios such as mobile phone simulation

ECS Flavor Naming Rules

Figure 1 shows a flavor name, which consists of the instance generation, instance size, and memory/vCPU ratio.

Figure 1 Flavor naming rule

Certain flavor names contain additional identifiers. For example, c6h.22xlarge.2.physical contains the additional identifier physical, which indicates that this BMS instance shares a resource pool with ECS instances.

  • Instance types

    The instance type is named in the following format: CPU architecture + instance family + instance generation + additional capabilities.

    Table 2 describes the naming rule for the instance type.
    Table 2 Naming rule for the instance type

    Item

    Description

    Rule

    Example

    Processor family

    Indicates the CPU architecture.

    Represented in a lowercase letter.

    • x86: There is no prefix by default.
    • Kunpeng: The prefix is a lowercase letter k.

    Instance family

    Indicates the typical scenarios.

    Represented in a lowercase letter.

    See Table 3.

    Instance generation

    Indicates the evolution of the instance family.

    Represented in a digit. The digit increases as the hardware and architecture evolves.

    None

    Additional capabilities

    Indicate the enhanced capabilities of the instance when compared with the other instances of the same generation.

    Represented in lowercase letters.

    See Table 4.

    Table 3 Instance families

    Application Scenarios

    Segmented Scenarios

    Instance Family

    Description

    General-purpose

    General Computing-Basic

    t

    Turbo

    General Computing

    s

    Standard

    General Computing-plus

    c

    Compute

    High-performance computing

    High-Performance Computing

    h

    High Performance

    Big data

    Disk-intensive

    d

    Disk

    Ultra-high I/O (large-capacity local disks)

    i

    IOPS

    Ultra-high I/O (small-capacity local disks)

    ir

    IOPS Raid

    Memory-intensive

    Memory-optimized

    m

    Memory

    Large-Memory

    e

    Enhanced Memory

    Computing-accelerated

    GPU computing-accelerated

    p

    Parallel

    GPU graphics-accelerated

    g

    Graphic

    GPU inference-accelerated

    pi

    Parallel Inference

    FPGA-accelerated

    fp

    FPGA Performance

    AI inference-accelerated

    ai

    Ascend Inference

    Table 4 Additional capabilities

    Suffix

    Example

    Description

    ne

    c3ne

    Network Enhanced

    s

    c6s

    Standard

    v

    p2v

    NVlink

    h

    c6h

    High performance

  • Instance sizes

    The instance sizes can be small, medium, large, xlarge, or Nxlarge, as shown in Table 5.

    For example, 2xlarge in s6.2xlarge.4 indicates that there are 8 vCPUs.
    Table 5 Mapping between instance size and the number of vCPUs

    Instance Size

    vCPUs

    small

    1

    medium

    1

    large

    2

    xlarge

    4

    Nxlarge

    N x 4. A larger value of N indicates more vCPUs.

  • Memory/vCPU ratio

    It is represented by a digit.

    For example, 4 in s6.2xlarge.4 indicates a memory-to-vCPU ratio of 4, which means that there are 8 vCPUs and the memory size is 32 GiB.

  • Additional identifies

    The BMSs that share a resource pool with ECSs are identified by physical.

    For example, physical in c6h.22xlarge.2.physical indicates a standard BMS instance that shares a resource pool with ECS instances.

vCPU

The processor uses the hyper-threading technology. The CPU exposes two execution contexts per physical core. This means that one physical core now works like two "logical cores" that can handle different software threads.

For example, a 10-core physical CPU contains 20 vCPUs (threads).

Network QoS

Network QoS uses basic technologies to improve the quality of network communication. A network with QoS enabled offers predictable network performance and effectively allocates network bandwidth to use network resources.

To obtain the QoS data of an ECS flavor, including the maximum bandwidth, assured bandwidth, maximum intranet PPS, NIC multi-queues, and maximum NICs, see A Summary List of x86 ECS Specifications.

The intranet bandwidth and PPS of an ECS are determined by the ECS flavor.
  • Assured intranet bandwidth: indicates the guaranteed bandwidth allocated to an ECS when there is a network bandwidth contention in the entire network.
  • Maximum intranet bandwidth: indicates the maximum bandwidth that can be allocated to an ECS when the ECS does not compete for network bandwidth (other ECSs on the host do not have high requirements on network bandwidth).
  • Maximum intranet PPS: indicates the maximum ECS capability in sending and receiving packets.

    PPS: packets per second, indicates the number of packets sent per second. It is usually used to measure the network performance.

  • NIC multi-queues: allocates NIC interrupt requests to multiple vCPUs for higher PPS performance and bandwidth
  • Maximum NICs: indicates the maximum number of NICs that can be attached to an ECS.
    • For instructions about how to test packet transmit and receive, see How Can I Test Network Performance?
    • For instructions about how to enable NIC multi-queue, see Enabling NIC Multi-Queue.
    • The maximum bandwidth is the total bandwidth allocated to an ECS. If an ECS has multiple NICs, the sum of the maximum bandwidths allocated to all NICs cannot exceed the maximum bandwidth allocated to the ECS.

Dedicated and Shared ECSs

Table 6 Differences between dedicated and shared ECSs

Dimension

Dedicated ECS

Shared ECS

CPU Allocation

CPUs are exclusively used and there is no CPU contention.

CPUs are shared and CPU contention may occur.

Feature

  • High performance
  • Dedicated and stable computing, storage, and network resources
  • High costs
  • Unstable performance when loads are high
  • Shared computing, storage, and network resources
  • Low costs

Application Scenario

For enterprises that have high requirements on service stability

For small- and medium-sized websites or individuals that have requirements on cost-effectiveness

ECS Specifications

Specifications except general computing and general computing-basic

x86 computing: