Fine-Grained Network Scheduling

Last updated: 2024-12-24 15:47:21
The fine-grained scheduling capability of network provides a series of features to ensure the service quality of business networks, improving network performance in all aspects and flexibly limiting the use of network by a container.
Feature 1: Throttling of Inbound and Outbound Directions
Overview
Limit the inbound and outbound bandwidth of the container.
Operation Steps
1. Deploy QoS Agent.
2. Within the Add-on Management page in the cluster, locate the successfully deployed QoS Agent, and click Update Configuration on the right.
3. On the add-on configuration page for modifying QoS Agent, tick the Network QoS Enhancement.
4. Click Complete.
5. Deploy a business.
6. Deploy the PodQOS object associated with the business, and select the business to apply, as shown below:
apiVersion: ensurance.crane.io/v1alpha1
kind: PodQOS
metadata:
  name: a
spec:
  labelSelector:
    matchLabels:
      k8s-app: a # select the Label of the business that needs to lower the priority
  resourceQOS:
    netIOQOS:
      netIOLimits:
        rxBps: 50 # rxBps represents the maximum inbound bandwidth, in Mbps. 0 means no limit. The maximum input value is a 13-digit integer.
        txBps: 50 # txBps represents the maximum outbound bandwidth, in Mbps. 0 means no limit. The maximum input value is a 13-digit integer.
Feature 2: Priority (Absolute Bandwidth Preemption)
Overview
Users can set different priorities for different containers and allocate bandwidth resources of network interfaces according to the priorities. An example of scheduling policy for inbound bandwidth with three priority levels is provided, which can be extended to more priority levels:
Containers with the highest priority (priority 0) are not subject to network QoS limitations and can use bandwidth resources freely.
When the total bandwidth used by all containers with the lower priorities (priorities 1 and 2) is less than rx_bps_max minus the bandwidth of the highest priority, and the bandwidth which can be used by each container can is not subject to limitations and can exceed rx_bps_min.
When the total bandwidth used by all containers with the lower priorities (priorities 1 and 2) exceeds rx_bps_max, the containers can use up to the difference between rx_bps_max and the total traffic of the containers with a higher priority. If the difference is less than rx_bps_min, then the containers with the lower priority are guaranteed to use the bandwidth of at least rx_bps_min.
The bandwidth of the container with the highest priority (priority 0) can exceed rx_bps_max/tx_bps_max, while the containers with lower priorities cannot exceed this parameter.
Scenarios
Scenario 1
When containers with higher priorities have lower bandwidth usage, an idle bandwidth will be allocated to the containers with low priorities.
When containers with higher priorities have higher bandwidth usage exceeding rx_bw_max, only the minimum guaranteed bandwidth will be allocated to the containers with low priorities.
﻿
Scenario 2
When containers with higher priorities have lower bandwidth usage, an idle bandwidth will be allocated to the containers with low priorities.
When a container with medium priority is added, the bandwidth of the containers with relatively low priorities will be preempted.
﻿
Scenario 3
When containers with higher priorities have lower bandwidth usage, an idle bandwidth will be allocated to the containers with low priorities.
When a container with medium priority is added, the bandwidth of the containers with relatively low priorities will be preempted. The maximum bandwidth which can be preempted shall subtract the bandwidth used by the containers with higher priorities and the minimum guaranteed bandwidth of the containers with lower priorities.
﻿
Operation Steps
1. Deploy QoS Agent.
2. Within the Add-on Management page in the cluster, locate the successfully deployed QoS Agent, and click Update Configuration on the right.
3. On the add-on configuration page for modifying QoS Agent, tick the Network QoS Enhancement.
4. Click Complete.
5. Deploy the business A.
6. Deploy the PodQOS object associated with the business, and select the business to apply, as shown below:
apiVersion: ensurance.crane.io/v1alpha1
kind: PodQOS
metadata:
  name: a
spec:
  labelSelector:
    matchLabels:
      k8s-app: a # select the Label of the business
  resourceQOS:
    netIOQOS:
      netIOPriority: 6 # select the network priority of 0-7, a total of 8 levels
7. Deploy the business B.
8. Deploy the PodQOS object associated with the business, and select the business to apply, as shown below:
apiVersion: ensurance.crane.io/v1alpha1
kind: PodQOS
metadata:
  name: b
spec:
  labelSelector:
    matchLabels:
      k8s-app: b # select the Label of the business
  resourceQOS:
    netIOQOS:
      netIOPriority: 7 # select the network priority of 0-7, a total of 8 levels
This feature needs to be used jointly with the node bandwidth limitation in NodeQOS, specifying rxBpsMin, rxBpsMax, txBpsMin and txBpsMax with NodeQOS.
rxBpsMin: The minimum guaranteed bandwidth in the inbound direction at the per-priority level. It is shared by the containers at this level. Currently, the minimum guaranteed bandwidth in the inbound direction is the same at different priority levels, except for priority 0.
txBpsMin: The minimum guaranteed bandwidth in the outbound direction at the per-priority level. It is shared by the containers at this level. Currently, the minimum guaranteed bandwidth in the outbound direction is the same at different priority levels, except for priority 0.
rxBpsMax: Maximum inbound bandwidth of the network interface.
txBpsMax: Maximum outbound bandwidth of the network interface.
Note:
Unit: Mbps.
apiVersion: ensurance.crane.io/v1alpha1
kind: NodeQOS
metadata:
  name: total
spec:
  netLimits:
    rxBpsMin: 10
    rxBpsMax: 10240
    txBpsMin: 10
    txBpsMax: 10240
Feature 3: Port Allowlist
Overview
To prevent the containers with low priorities from starving, there are two mechanisms:
1. Minimum guaranteed bandwidth of per-priority: Shared by the containers with this priority.
2. Port allowlist mechanism: Users can add a local port or peer port from the container to the allowlist. The traffic of this port will not be throttled. It is generally used to protect control messages of specific protocols.
Operation Steps
1. Deploy QoS Agent.
2. Within the Add-on Management page in the cluster, locate the successfully deployed QoS Agent, and click Update Configuration on the right.
3. On the add-on configuration page for modifying QoS Agent, tick the Network QoS Enhancement.
4. Click Complete.
5. Deploy a business.
6. Deploy the PodQOS object associated with the business, and select the business to apply, as shown below:
apiVersion: ensurance.crane.io/v1alpha1
kind: PodQOS
metadata:
  name: a
spec:
  labelSelector:
    matchLabels:
      k8s-app: a # select the Label of the business
  resourceQOS:
    netIOQOS:
      whitelistPorts: # The local port 5201 and the remote port 5205 representing the business a are set as allowlist ports and are not throttled. The default values for lports and rports are 0, with the value range of 0- to 13-digit integers.
        lports: 5201
        rports: 5205
﻿
﻿
tencent cloud

Recent Pages

Fine-Grained Network Scheduling

Feature 1: Throttling of Inbound and Outbound Directions

Overview

Operation Steps

Feature 2: Priority (Absolute Bandwidth Preemption)

Overview

Scenarios

Scenario 1

Scenario 2

Scenario 3

Operation Steps

Feature 3: Port Allowlist

Overview

Operation Steps

Was this page helpful?

Was this page helpful?

tencent cloud

Sign Up

Log in

Recent Pages

Fine-Grained Network Scheduling

Feature 1: Throttling of Inbound and Outbound Directions

Overview

Operation Steps

Feature 2: Priority (Absolute Bandwidth Preemption)

Overview

Scenarios

Scenario 1

Scenario 2

Scenario 3

Operation Steps

Feature 3: Port Allowlist

Overview

Operation Steps

Was this page helpful?

Was this page helpful?
