Distributed flow control

Patent metadata

Field	Value
Publication number	US-10447789-B2
Application number	US-201615169897-A
Country	US
Kind code	B2
Filing date	Jun 1, 2016
Priority date	Dec 31, 2013
Publication date	Oct 15, 2019
Grant date	Oct 15, 2019

Abstract

A management server may provide distribute flow control to a computer server system for providing online services to users. The computer server system includes a bus cluster, a plurality of caller systems, and a plurality of callee clusters. The management server may be configured to receive from the bus cluster health condition data of a callee cluster of the plurality of callee cluster in a first operation cycle; determine whether the callee cluster is slow or abnormal based on the health condition data in the first cycle; adjust a data flow limit in the bus cluster in a second cycle after the first cycle according to whether the callee cluster is slow or abnormal in the first cycle; and direct the bus cluster to control the data flow of the computer servers in the bus cluster in the second cycle according to the adjusted data flow limit.

First claim

The invention claimed is:

1. A management server for providing distributed flow control to a computer server system that provides online services to users, the computer server system comprising a bus cluster, a caller system, and a plurality of callee clusters, the management server comprising:
a memory storing a set of instructions; and
a processor in communication with the memory, wherein, when executing the set of instructions, the processor is programmed to:
receive, from the bus cluster, health condition data of a callee cluster of the plurality of callee clusters in a first operation cycle of the computer server system;
determine, based on the health condition data, whether the callee cluster is slow or abnormal in the first operation cycle according to time consumption data of the callee cluster;
in response to determining whether the callee cluster is slow or abnormal in the first operation cycle, adjust a data flow limit in the bus cluster in a second operation cycle of the computer server system, wherein the second operation cycle is after the first operation cycle;
direct, according to the adjusted data flow limit, the bus cluster to control the data flow of a computer server in the bus cluster in the second operation cycle;
wherein the data flow limit is an upper data flow limit for a computer server in the bus cluster in the second operation cycle;
wherein, when the processor is programmed to adjust the data flow limit, the processor is programmed to:
in response to determining whether the callee cluster is slow or abnormal in the first operation cycle, adjust a capacity of a computer server in the callee cluster, and
determine the upper data flow limit for the computer server in the bus cluster in the second operation cycle according to the adjusted capacity;

wherein, when the processor is programmed to determine the upper data flow limit for the computer server in the bus cluster, the processor is programmed to:
determine a first value being a product of the adjusted capacity multiplied by a fraction of a data flow of the computer server in the bus cluster in the first operation cycle divided by a total data flow of all the computer servers in the bus cluster in the first operation cycle,
determine a second value being a fraction of the adjusted capacity divided by a total number of the computer servers in the bus cluster, and
take a larger value between the first value and the second value as the upper data flow limit for the computer server in the bus cluster in the second operation cycle; and

wherein the processor is further programmed to receive from an administrator an instruction to conduct a pressure test targeting a callee cluster as a pressure test target, and in response to receiving the instruction to conduct the pressure test, the processor is programmed to:
send a short cycle reporting instruction to the bus cluster to report the health condition data according to a short cycle, wherein the short cycle has a shorter time period than the first operation cycle,
receive the health condition data reported by the bus cluster according to the short cycle,
control the data flow in the bus cluster according to the short cycle, and
reduce an upper limits of request amount of the plurality of callee clusters except for the pressure test target by a first default value in each short cycle.

2. The management server of claim 1 , wherein to adjust the capacity of the callee cluster, the processor is further programmed to:
reduce the capacity of the callee cluster when the callee cluster is slow or operates abnormally; and
increase the capacity of the callee cluster when the callee cluster is not slow or operates normally.

3. The management server of claim 2 , wherein to reduce the capacity the processor is further programmed to reduce the capacity by a default percentage of an initial value of the capacity; and
to increase the capacity the processor is further programmed to increase the capacity by the default percentage of the initial value of the capacity.

4. The management server of claim 1 , wherein to direct the bus cluster to control the data flow of the computer servers in the bus cluster in the second operation cycle the processor is further programmed to:
obtain a total data flow of the computer servers in the bus cluster in the second operation cycle; and
when the total data flow exceeds a preset times of the adjusted capacity of the callee cluster, send an alert to an administrator of the management server.

5. The management server of claim 1 , wherein the health condition data comprises at least one of an average time consumption of the callee cluster spending on requests in a reporting cycle, distribution data of time consumptions of the callee cluster spending on requests in the reporting cycle, and an overtime rate of the time consumptions of the callee cluster; and
to determine whether the callee cluster is slow or abnormal in the first operation cycle the processor is further programmed to determine that the callee cluster is slow when the callee cluster in the first operation cycle meets at least one condition of:
the average time consumption of the callee cluster reaches a first threshold value;
among the time consumption distribution data of the callee cluster, a request ratio in each time consumption interval is not in a default range;
the overtime rate of the callee cluster reaches a second threshold value.

6. The management server of claim 1 , wherein the health condition data comprises a success rate and an error rate; and
to determine whether callee cluster is slow or abnormal the processor is further programmed to determine that the callee cluster is abnormal when the callee cluster in the first operation cycle meets at least one condition of:
the success rate of the callee cluster is lower than a third threshold value;
the error rate of the callee cluster reaches a forth threshold value.

7. The management server of claim 1 , wherein:
when the pressure test meets at least one condition that: a request amount of all callee clusters except for the pressure test target decreases to a second default value or the pressure test target is slow or abnormal, the processor is further programmed to:
stop the pressure test;
recover a data flow limit data before the pressure test; and
send a normal cycle reporting instruction to the bus cluster.

8. A method for providing distributed flow control to a computer server system, comprising:
providing the computer server system configured to provide online services to users, wherein the computer server system comprises a bus cluster including a plurality of computer servers, a caller system including a plurality of computer servers, a plurality of callee clusters including a plurality of computer servers, and a management daemon server;
receiving, by the management daemon server from the bus cluster, health condition data of a callee cluster of the plurality of callee clusters in a first operation cycle of the computer server system;
determining, by the management daemon server, whether the callee cluster is slow or abnormal based on the health condition data in the first operation cycle according to time consumption data of the callee cluster;
adjusting, by the management daemon server, a data flow limit of the bus cluster in a second operation cycle of the computer server system after the first operation cycle according to whether the callee cluster is slow or abnormal in the first operation cycle;
directing, by the management daemon server, the bus cluster to control the data flow of the computer servers in the bus cluster in the second operation cycle according to the adjusted data flow limit;
wherein the data flow limit is a

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Song Yinbo
• Guo Rui
• Mai Maoshi
• Niu Zhiqiang
• Zhang Jinsong
• Cheng Wei
• Liu Haoran
• Jiang Qiang
• Ren Bo
• Zhou Baojun

Classifications

• H04L67/42

  Electricity · mapped topic

• H04L41/0893

  Assignment of logical groups to network elements · CPC title

• H04L12/40032 Primary

  Details regarding a bus interface enhancer · CPC title

• H04L67/16 Primary

  Electricity · mapped topic

• H04L47/10

  Flow control; Congestion control · CPC title