This feature is in invitation-only preview. To use it, contact your Alibaba Cloud account manager.
Basic QoS allocates bandwidth proportionally but cannot guarantee low latency for critical services during congestion. Priority-based QoS on Transit Router (TR) inter-region connections combines strict priority preemption and weighted fair sharing, and lets you assign different services to links of varying quality.
How it works
Priority-based QoS attaches a QoS policy to an inter-region connection. The policy contains multiple queues at different priorities. Traffic is matched to a queue based on the Differentiated Services Code Point (DSCP) value in the packet header. Each queue determines how it preempts bandwidth and which link type it uses.
Transit Router (TR)
└─ Attachment (inter-region connection)
└─ InterRegionTrafficQosPolicy (QoS policy)
├─ Queue (Matches by DSCP, determines queue type and link level)
├─ Queue
└─ ... (Up to 8)
Queue scheduling mode
Unlike basic QoS, priority-based QoS supports unlimited preemption.
|
Mode |
Basic Rate Limit |
Priority Scheduling |
|
Preemption |
No. All queues share bandwidth equally. |
Yes. The Priority Pass has unlimited preemption. |
|
Allowed queue types |
Basic Queue |
Priority Pass + Equal Share |
|
Queue bandwidth parameters |
By Percentage or By Absolute Value |
After accounting for the bandwidth used by the Priority Queue, the remaining bandwidth can only be allocated by percentage. |
|
Use cases |
General service isolation, compliance-based rate limiting |
Financial trading, low-latency assurance |
For more information about the traffic scheduling policy matching process and the differences between DSCP values in policies and rules, see basic QoS.
Queue types (Priority Scheduling only)
-
Priority Pass (PQ): A QoS policy allows a maximum of one PQ. Traffic in this queue has unlimited preemption. As long as there are packets in the PQ, they are forwarded immediately, and other queues must yield.
-
Equal Share (WFQ): Multiple WFQs are allowed. These queues share the bandwidth left over from the PQ based on percentages. The system automatically normalizes the weights, so the sum is not required to equal 100.
Link Type (Priority Scheduling only)
-
Link type selection is available only when the inter-region connection uses the Pay-By-Data-Transfer billing method. The available options are Gold and Platinum.
-
Unmarked traffic is automatically directed to the Default Queue, which inherits the link type of the inter-region connection.
-
The link type determines the forwarding plane for queue traffic within the Alibaba Cloud network. All service levels use dedicated private lines and do not traverse the public internet.
Level
Description
Typical use cases
Platinum
Highest SLA dedicated line (e.g., a dedicated channel for financial cloud)
Core transaction orders
Gold
High-assurance internal network link
Real-time risk control, market data feeds
Default Line Type
Inherits the link type of the inter-region connection.
Other general-purpose traffic
Note: The link type only determines the path selection and does not affect preemption between queues. Preemption is determined solely by the queue type (PQ or WFQ).
Use cases
Scenario 1: High-frequency trading
-
Requirement: Order packets (DSCP=46) must have a latency of less than 5 ms and must not be preempted by traffic from other services, such as market data or logs.
-
Solution: Direct traffic with DSCP=46 to a PQ + Platinum link. Route market data and log traffic to a WFQ + Gold link.
Scenario 2: Cross-region ERP
-
Requirement: Database Redo logs require guaranteed bandwidth and low latency. File backups can elastically use idle bandwidth.
-
Solution: Direct Redo logs to a high-weight WFQ + Platinum link. Direct backup traffic to a low-weight WFQ + Default Line Type. If strict preemption is not required, you do not need to configure a PQ.
Scenario 3: Cross-region gaming
-
Requirement: Player commands (high DSCP value) must have zero jitter. Chat and announcement traffic should share the remaining bandwidth based on their weights.
-
Solution: Direct player commands to a PQ + Platinum link. Direct chat and announcement traffic to multiple WFQs that are allocated bandwidth based on their weights.
Mode selection
-
If you only need to enforce strict bandwidth partitioning, use Basic Rate Limit.
-
If you have critical services that require guaranteed low latency and cannot tolerate packet loss, use Priority Scheduling.
Create a priority-based QoS policy
Before you begin, make sure the following requirements are met:
-
The Transit Router (TR) that hosts the inter-region connection must be an Enterprise Edition instance.
-
The state of the inter-region connection is Attached.
-
You have used a traffic marking policy to DSCP for the target traffic.
Follow these steps:
-
In the QoS Policy column of the target inter-region connection, click Set.
-
For Queue Scheduling Mode, select Priority Scheduling. Then, Add Queue as needed. You can add up to eight queues.
We recommend configuring a PQ for critical services and WFQs for non-preemptive services.
-
Configure each queue:
-
Queue Type: For more information, see Queue types (Priority Scheduling only).
-
Link Type: You can select a link type only when the inter-region bandwidth billing method is Pay-By-Data-Transfer. For more information, see Link types (Priority Scheduling only).
-
Matching DSCP: Matches the DSCP value assigned to traffic by a traffic marking policy, as specified in the DSCP setting. Traffic matching the specified DSCP value is classified into the current queue. A single queue can match multiple comma-separated DSCP values.
-
Guaranteed Bandwidth: After excluding bandwidth used by the Priority Pass, the remaining bandwidth is allocated by percentage. When the Priority Pass is fully utilized, the actual bandwidth available to other queues is reduced, and no minimum bandwidth is guaranteed.
-
More information
Limits
-
Only Enterprise Edition TRs support priority-based QoS.
-
You can change the Link Type for a queue in Priority Scheduling mode depending on the Bandwidth Allocation Mode you selected when you created the inter-region connection:
-
If you select Pay-By-Data-Transfer, you can change the link type.
-
If you select Allocate From Bandwidth Plan, you cannot change the link type. The link type for the queue defaults to Gold.
-
-
Maximum number of queues per QoS policy: 8 (including the default queue).
-
Maximum number of Priority Pass per policy: 1.
-
Maximum number of QoS policies per inter-region connection: 1.
-
DSCP value range: 0 to 63.
Billing
Priority-based QoS itself is free of charge. You are charged a cross-region bandwidth fee based on the link type and the standard TR billing rules. If you change the link type within an hour, you are billed for the highest service level used during that hour (Levels: Platinum > Gold). A downgrade takes effect in the next billing hour.
FAQ
Migrating from basic QoS to priority-based QoS?
The scheduling mode cannot be changed after a policy is created. You must delete and recreate the policy. Record the original configuration, delete all queues, delete the policy, and then recreate it in Priority Scheduling mode. QoS is temporarily inactive on the connection during this process, so we recommend performing this operation during off-peak hours.
Does the link type determine preemption?
No. The link type only controls the physical path traffic takes, affecting transmission quality and cost. Queue preemption is determined solely by the Queue Type. For example, a Priority Pass on a Gold link can still preempt a Equal Share on a Platinum link. Although they use different physical paths, preemption is enforced at the bandwidth scheduling layer.