Load Balancer
Topic | Explanation | Application scenario |
Server type | Support adding the following resources as backend servers:
warning When the NLB instance enables client address is maintained in a public network client access scenario, binding an EIP to a server-type backend server causes traffic for that IP address to be unreachable. |
|
IP type | Supports IPv4 addresses as backend servers as long as the network is reachable. e.g.
|
|
Topic | Explanation | Limits |
TCP | NLB forwards the access requests to the backend server through TCP protocol. | Only support attaching TCP and TLS listeners |
UDP | NLB forwards the access requests to the backend server through UDP protocol. | Only support attaching UDP listeners |
Scheduling algorithm | Explanation | Application scenario |
Weighted Round Robin (WRR) | This algorithm distributes access requests from users to different backend servers in sequence based on the weight of each backend server. The better the performance of a server, the higher the weight. Backend servers with higher weights receive polling more frequently (higher probability). | Applicable to short connection services. This algorithm is recommended when each request consumes roughly the same processing time on the backend server, handles the same or similar request types, and request times vary little. |
Weighted Least Connections (WLC) | First calculate the value of current connections/weight on the backend server, then distribute the requests to the backend server with the smallest value. This algorithm requires tracking the current connection count of each server. When a request is assigned to a backend server, its connection count increases by one; when the request connection terminates, times out, or completes, the connection count decreases by one. | WLC is applicable to persistent connection services, such as FTP. This algorithm is recommended when handling different requests with significantly varying processing times. |
Source IP Hash (SH) | Based on source IP address consistency hashing, requests from the same source address are mapped to the same backend server via a hash table. When removing or adding a backend server, it minimizes changes to the mapping between existing service requests and processing servers as much as possible. | Applicable to protocols without cookie functionality. Use this algorithm when requests from the same user need to be forwarded to the same backend server. |