All Products
Search

This Product

    Elastic Compute Service:Overview of the network bandwidth of ECS instances

    Document Center

    Elastic Compute Service:Overview of the network bandwidth of ECS instances

    Last Updated:Nov 12, 2024

    Network bandwidth is the maximum amount of data that can be transferred over a network in a specific period of time. Higher network bandwidth allows the transfer of a larger amount of data in the same period of time. Network bandwidth is classified into public bandwidth and internal bandwidth.

    Public bandwidth

    Public bandwidth is used to transfer data between Elastic Compute Service (ECS) instances and the Internet. Public bandwidth is classified into outbound public bandwidth and inbound public bandwidth. Outbound public bandwidth is used for traffic from ECS instances to the Internet. Inbound public bandwidth is used for traffic from the Internet to ECS instances. To enable public bandwidth for an ECS instance, you can assign a public IP address to the instance during instance creation or modify the instance bandwith configurations after instance creation. For more information, see Enable public bandwidth for an ECS instance and Modify bandwidth configurations. Alternatively, you can associate an elastic IP address (EIP) with the instance. For more information, see Associate one or more EIPs with an instance. You can also assign an IPv6 address to an ECS instance and enable IPv6 public bandwidth for the instance. For more information, see Step 3: Enable IPv6 public bandwidth for a Windows instance or Step 3: Enable IPv6 public bandwidth for a Linux instance.

    You are charged for outbound public bandwidth. The maximum inbound public bandwidth varies based on the outbound public bandwidth: If the outbound public bandwidth is less than or equal to 10 Mbit/s, the inbound public bandwidth is capped at 10 Mbit/s. If the outbound public bandwidth is greater than 10 Mbit/s, the inbound public bandwidth is equal to the outbound public bandwidth. If the outbound public bandwidth is greater than 1 Gbit/s, distributed throttling is automatically enabled. The maximum bandwidth value that you specify is evenly allocated to multiple ECS instances. For each ECS instance, public bandwidth for single-flow traffic is limited to the specified maximum bandwidth value divided by the number of instances, and public bandwidth for multi-flow traffic is limited to the specified maximum bandwidth value.

    You can use Anycast EIPs to improve Internet access based on the stable Border Gateway Protocol (BGP) lines and the global transmission network of Alibaba Cloud. For information about Anycast EIPs, see What is Anycast EIP?

    BGP (Multi-ISP) is suitable for instance communication over the Internet in the China (Hong Kong) region and between the China (Hong Kong) region and other regions outside the Chinese mainland. If you use BGP (Multi-ISP) lines to connect instances located in the China (Hong Kong) region with instances located in regions in the Chinese mainland, international ISP services are used. To reduce network latency in this scenario, select BGP (Multi-ISP) Pro.

    Bandwidth billing

    Public bandwidth supports the pay-by-bandwidth and pay-by-traffic metering methods. For more information, see Public bandwidth. To share and reuse network bandwidth in a region, create an Internet Shared Bandwidth instance in the region. For information about Internet Shared Bandwidth, see What is an Internet Shared Bandwidth? You can associate EIPs with an Internet Shared Bandwidth instance that resides in the same region as the EIPs. This way, you can reuse network bandwidth on the Internet Shared Bandwidth instance to reduce costs. You can apply data transfer plans to the IPv4 data transfer of eligible resources to reduce the costs of your public bandwidth usage. Eligible resources include public IP addresses, EIPs, Classic Load Balancer (CLB) instances, and Internet Shared Bandwidth instances that use the pay-by-traffic (also known as pay-by-data-transfer) metering method. You cannot apply data transfer plans to EIPs of BGP (Multi-ISP) Pro. For more information about data transfer plans, see What is a data transfer plan?

    Bandwidth security

    • By default, Alibaba Cloud Security Center provides a DDoS mitigation capacity for each ECS instance free of charge. The mitigation capacity varies based on the instance type and can be up to 5 Gbit/s. For more information, see View the thresholds that trigger blackhole filtering in Anti-DDoS Basic.

    • After you activate Anti-DDoS Basic, Alibaba Cloud Security Center monitors inbound traffic to ECS instances in real time. When an ultra-large amount of traffic or suspicious traffic such as DDoS attack traffic is detected, Security Center redirects the traffic from the intended paths to a scrubbing device. The scrubbing device identifies and removes malicious traffic, and then returns legitimate traffic. Then, the legitimate traffic is forwarded to ECS instances by using the intended paths. For more information, see What is Anti-DDoS Origin?

    • When an ECS instance is under a DDoS attack, you can defend against the attack at the earliest opportunity based on the pushed event. For more information, see Instance security events.

    Limits

    Starting November 27, 2020, the maximum bandwidth value available for new or updated ECS instances vary based on the throttling policies of your account. To apply for an increase in bandwidth quotas, submit a ticket.

    The following throttling policies apply:

    • In each region, the total maximum bandwidth of all ECS instances that use the pay-by-traffic metering method cannot exceed 5 Gbit/s.

    • In each region, the total maximum bandwidth of all ECS instances that use the pay-by-bandwidth metering method cannot exceed 50 Gbit/s.

    For more information, see the Public bandwidth limits section of the "Limits" topic.

    Internal bandwidth

    Internal bandwidth is used to transfer data between ECS instances over the internal network in the same virtual private cloud (VPC) and region. You can connect ECS instances to ApsaraDB RDS instances, Server Load Balancer (SLB) instances, and Object Storage Service (OSS) buckets over the internal network. You are not charged for in-region data transfer over the internal network. The internal bandwidth value varies based on the instance type. For information about the internal bandwidth value that is supported by each instance type, see Overview of instance families. Internal bandwidth is allocated to each ECS instance. If multiple network interfaces are bound to an ECS instance, the total internal bandwidth that is used by the network interfaces cannot exceed the internal bandwidth of the instance.

    Note

    • Cross-zone internal bandwidth varies based on the bandwidth specifications of instance types. Network latency increases with distance between zones.

    • Internal bandwidth in a deployment set or across deployment sets also varies based on the bandwidth specifications of instance types.

    • In most cases, the network bandwidth performance of an instance type is defined based on the transmission of data packets that are 1,514 bytes in size. If the size of data packets transmitted on an instance is less than 1,514 bytes, the achievable network bandwidth performance decreases.

    When you use internal bandwidth, take note of the following items:

    • Physical network bandwidth is shared across ECS instances. The internal bandwidth of an ECS instance may be affected by the internal bandwidth usage of other ECS instances. In most cases, an ECS instance can achieve the internal bandwidth provided by the instance type. For information about how to test internal bandwidth performance, see Test the network performance of an instance.

    • If your business occasionally requires a network bandwidth that exceeds the baseline bandwidth of ECS instances, you can select an instance type that supports burst bandwidth.

    • In scenarios such as high-performance computing, big data processing, and AI training, you can select an Elastic Remote Direct Memory Access (eRDMA)-capable instance type to provide low-latency, high-throughput network services. RDMA transfers data from user-mode programs to Host Channel Adapters (HCAs) for network transmission, without involving the kernel stack. RDMA helps greatly reduce CPU load and latency.

    • To achieve internal bandwidth of 100 Gbit/s or higher per instance, select an instance type that supports network card mappings and specify network card indexes to bind elastic network interfaces (ENIs) to different network cards at the underlying layer. This helps maximize bandwidth utilization. For information about network card mappings, see the Basics section of the "Overview" topic.

    • If your business requires concurrent data transmission and reception, we recommend that you use seventh-generation or later instance types to achieve full-duplex transmit and receive bandwidth. The transmission and reception rates are separately calculated. Data can be transmitted and received at the full rate at the same time.

    Burst bandwidth

    Specific instance types in sixth-generation or later instance families support network burst bandwidth. When an ECS instance of these instance types has sufficient network bandwidth resources and available network burst credits, the instance can consume network burst credits to burst beyond its network baseline and achieve a higher transmission speed in response to traffic spikes. Network burst bandwidth is a way to utilize idle resources. No service level agreement (SLA) commitments are made for network burst bandwidth.

    For information about the baseline bandwidth, burstable network bandwidth capabilities, and maximum burst bandwidth of an instance type, see Overview of instance families.

    Network burst credits

    An ECS instance can accrue credits when the instance uses network bandwidth that is lower than its baseline bandwidth in the internal network. The maximum number of network burst credits that an instance can accrue varies based on the instance type. When an ECS instance requires network bandwidth that is higher than its baseline bandwidth, the instance consumes credits to temporarily burst the network bandwidth beyond the baseline bandwidth. When the network burst credits of the ECS instance are exhausted, the instance returns to its baseline bandwidth.

    Detailed rules for network burst credits

    • Credits

      Credits determine how long an ECS instance can burst its network bandwidth beyond the baseline bandwidth. The burst bandwidth is inversely proportional to the duration. The higher the burst bandwidth, the shorter the duration. The total amount of bandwidth provided by a credit is fixed. For example, an ECS instance of the ecs.g8i.large instance type supports up to 15 Gbit/s of burst bandwidth and can consume one credit to do the following:

      • Use 15 Gbit/s of bandwidth for 1 second.

      • Use 7.5 Gbit/s of bandwidth for 2 seconds.

      • Use 5 Gbit/s of bandwidth for 3 seconds.

    • Launch credits

      When you create and start an ECS instance, the instance receives the maximum number of credits as its launch credits.

    • Maximum number of credits

      The maximum number of credits that an ECS instance can accrue. An ECS instance of a higher-specification instance type can accrue a larger maximum number of credits.

    • Earn credits

      An ECS instance earns credits when the instance is running without generating network traffic. The speed at which an ECS instance earns credits varies based on the instance type and is calculated by using the following formula: (Baseline bandwidth/Maximum burst bandwidth) × 60 seconds. For example, an ECS instance of the ecs.g8i.large instance type earns 10 credits per minute, which is calculated by using the following formula:

      (2.5 Gbit/s/15 Gbit/s) × 60 seconds = 10.

    • Accrue credits

      A running ECS instance can accrue credits when the instance uses network bandwidth that is lower than its baseline bandwidth. A limit is imposed on the number of credits that an ECS instance can accrue. The speed at which an ECS instance accrues credits varies based on the bandwidth usage of the instance. The higher the bandwidth usage, the slower the accrual speed.

      The number of credits that an ECS instance can accrue per minute is calculated by using the following formula: (Baseline bandwidth - Used bandwidth)/Maximum burst bandwidth × 60 seconds. For example, an ECS instance of the ecs.g8i.large instance type has a baseline bandwidth of 2.5 Gbit/s and can burst its network bandwidth to up to 15 Gbit/s.

      • Each minute the ECS instance uses a network bandwidth of 2.0 Gbit/s, the instance accrues 2 credits: (2.5 Gbit/s - 2.0 Gbit/s)/15 Gbit/s × 60 seconds = 2.

      • Each minute the ECS instance uses a network bandwidth of 1.5 Gbit/s, the instance accrues 4 credits: (2.5 Gbit/s - 1.5 Gbit/s)/15 Gbit/s × 60 seconds = 4.

    • Consume credits

      An ECS instance consumes credits when the instance uses network bandwidth that is higher than its baseline bandwidth. When the ECS instance exhausts its credits, the network bandwidth of the instance is capped at the baseline bandwidth. The speed at which an ECS instance consumes credits varies based on the bandwidth usage. The higher the bandwidth usage, the faster the consumption speed.

      The number of credits that an ECS instance can consume per minute is calculated by using the following formula: (Used bandwidth/Maximum burst bandwidth) × 60 seconds. For example, an ECS instance of the ecs.g8i.large instance type can burst its network bandwidth to up to 15 Gbit/s.

      • Each minute the ECS instance uses a network bandwidth of 15 Gbit/s, the instance consumes 60 credits: 15 Gbit/s/15 Gbit/s × 60 seconds = 60.

      • Each minute the ECS instance uses a network bandwidth of 10 Gbit/s, the instance consumes 40 credits: 10 Gbit/s/15 Gbit/s × 60 seconds = 40.

    Monitor network bandwidth

    You can use CloudMonitor to monitor network bandwidth.

    References