[figure data] how long does it take for the equity network to penetrate 1000 layers?

Here’s the table of contents:

Figure data relation path penetration test

The test runs on the overall scale of 11 Billion data sets , The amount of equity network data exceeds ten million , The test method is to start from a company and penetrate the equity network upwards 100 Layer and the 1000 layer , Return a path result 【LIMIT 1】. A hundred layers of penetration is equivalent to 101 Zhang MySQL The data sheet of 100 Of JOIN, A thousand layers of penetration is equivalent to 1001 Zhang MySQL The data sheet of 1000 Of JOIN,MySQL This query cannot be implemented .

Figure database selection

Figure database is ONgDB【 Technical communication QQ Group ：1061594137】, The deployment mode is three node causal cluster deployment , Two CORE Node and a REPLICA node . Service configuration is 64G Memory 8 nucleus CPU.

Figure data model description

• Data scale

The entire gallery contains nodes 1157523466 individual 【 1.1 billion nodes 】,1081043246 Relationships 【 Billion relationships 】. Equity network is modeled as isomorphic graph , The node is 16001402 individual 【 16 million 】, The relationship is 13980621 ten thousand 【 Fourteen million 】.

• Data modeling method

The equity network entity is labeled HORGShareHoldV002, The relationship is holding , The detailed data of shareholding is used shareholding_detail Fields store time series fields 【 Data modeling of time series can refer to integration ES Realization 】.

The equity network penetrates 100 floors

Through a hundred layers of query statements

Use postman perform cypher Inquire about

MATCH p=(n:HORGShareHoldV002)<-[r:` holding `*100]-(m) WHERE n.name=' Tianjin Qinling Business Management Co., Ltd ' RETURN p LIMIT 1

{
    "statements": [
        {
            "statement": "MATCH p=(n:HORGShareHoldV002)<-[r:` holding `*100]-(m) WHERE n.name=' Tianjin Qinling Business Management Co., Ltd ' RETURN p LIMIT 1",
            "resultDataContents": [
              "graph"
            ]
        }
    ]
}

Through a hundred floors 10 Test execution results

The equity network penetrates 1000 layers

Through a thousand layers of query statements

Use postman perform cypher Inquire about

MATCH p=(n:HORGShareHoldV002)<-[r:` holding `*1000]-(m) WHERE n.name=' Tianjin Qinling Business Management Co., Ltd ' RETURN p LIMIT 1

{
    "statements": [
        {
            "statement": "MATCH p=(n:HORGShareHoldV002)<-[r:` holding `*1000]-(m) WHERE n.name=' Tianjin Qinling Business Management Co., Ltd ' RETURN n,m LIMIT 1",
            "resultDataContents": [
              "graph"
            ]
        }
    ]
}

Through a thousand layers 10 Test execution results

Test result statistics

In theory, the algorithm time complexity of graph database path penetration is O(1),ONgDB【Neo4j】 The path penetration performance of is better than Nebula Graph Even distribution chart database has better performance 【 There is no test on the distributed graph database, but basic judgment can be made from the database architecture 】, Because the biggest performance consumption of distributed system lies in network communication , If it is an algorithm model related to artificial intelligence , The performance of centralized graph data architecture will be better 【 For a large number of graph data models, integration can be considered ES Store time series metrics , Solve the problem of data volume 】.

• Through a hundred floors

 The average time taken ：102 millisecond 
 Minimum time ：92 millisecond 
 Most time consuming ：129 millisecond 

• Through a thousand layers

 The average time taken ：29.3 second 
 Minimum time ：21.08 second 
 Most time consuming ：42.98 second