Test Environment

• Region: CN North-Beijing4
• AZ: AZ2
• Elastic Cloud Server (ECS): c3.2xlarge.2 flavor with 8 vCPUs, 16 GB of memory, and CentOS 7.6 (x86, 64 bit).
• DB engine versions of the cluster and replica set instances to be tested: 3.4 and 4.0.
• Each cluster instance includes two shard nodes.
• Specifications of the cluster and replica set Enhanced II instances to be tested: All the specifications supported by the instances are covered. For details, see Table 1 and Table 2.

  Table 1 Cluster instance specifications

  ID	DB Version	shard Class
  Cluster 1	3.4	2 vCPUs | 4 GB
  Cluster 2	3.4	2 vCPUs | 8 GB
  Cluster 3	3.4	4 vCPUs | 8 GB
  Cluster 4	3.4	4 vCPUs | 16 GB
  Cluster 5	3.4	8 vCPUs | 16 GB
  Cluster 6	3.4	8 vCPUs | 32 GB
  Cluster 7	3.4	16 vCPUs | 32 GB
  Cluster 8	3.4	16 vCPUs | 64 GB
  Cluster 9	3.4	32 vCPUs | 128 GB
  Cluster 10	3.4	64 vCPUs | 256 GB
  Cluster 11	4.0	2 vCPUs | 4 GB
  Cluster 12	4.0	2 vCPUs | 8 GB
  Cluster 13	4.0	4 vCPUs | 8 GB
  Cluster 14	4.0	4 vCPUs | 16 GB
  Cluster 15	4.0	8 vCPUs | 16 GB
  Cluster 16	4.0	8 vCPUs | 32 GB
  Cluster 17	4.0	16 vCPUs | 32 GB
  Cluster 18	4.0	16 vCPUs | 64 GB
  Cluster 19	4.0	32 vCPUs | 128 GB
  Cluster 20	4.0	64 vCPUs | 256 GB

  Table 2 Replica set specifications

  ID	DB Version	DB Instance Class
  Replica set 1	3.4	2 vCPUs | 4 GB
  Replica set 2	3.4	2 vCPUs | 8 GB
  Replica set 3	3.4	4 vCPUs | 8 GB
  Replica set 4	3.4	4 vCPUs | 16 GB
  Replica set 5	3.4	8 vCPUs | 16 GB
  Replica set 6	3.4	8 vCPUs | 32 GB
  Replica set 7	3.4	16 vCPUs | 32 GB
  Replica set 8	3.4	16 vCPUs | 64 GB
  Replica set 9	3.4	32 vCPUs | 128 GB
  Replica set 10	3.4	64 vCPUs | 256 GB
  Replica set 11	4.0	2 vCPUs | 4 GB
  Replica set 12	4.0	2 vCPUs | 8 GB
  Replica set 13	4.0	4 vCPUs | 8 GB
  Replica set 14	4.0	4 vCPUs | 16 GB
  Replica set 15	4.0	8 vCPUs | 16 GB
  Replica set 16	4.0	8 vCPUs | 32 GB
  Replica set 17	4.0	16 vCPUs | 32 GB
  Replica set 18	4.0	16 vCPUs | 64 GB
  Replica set 19	4.0	32 vCPUs | 128 GB
  Replica set 20	4.0	64 vCPUs | 256 GB

Test Tool

YCSB is an open-source database performance test tool. In this test, YCSB 0.12.0 is used.

For details on how to use this tool, see YCSB.

Test Metrics

Operations per Second (OPS): number of operations executed per second by a database

Test Procedure

1. Configure the workload configuration file.

   Set the values of readproportion, insertproportion, and updateproportion in the workload file by referring to Table 3.

   Set the value of recordcount in the workload file by referring to Table 4.

   Example: Configure the workload_s1 file.

   • recordcount = 100000000
   • operationcount = 100000000
   • insertproportion = 1
   • readproportion = 0
   • updateproportion = 0
   • scanproportion = 0
   

   The values of recordcount and operationcount are the same.

   The sum of the values of insertproportion, readproportion, updateproportion, and scanproportion is 1.

2. Run the following command to create a test database table on the DB instance, for example, usertable:

   ./bin/mongo --host $mongosIP --port $port << END_CMD

   use admin

   db.auth('rwuser', '$pwd')

   use ycsb

   sh.enableSharding("ycsb")

   sh.shardCollection("ycsb.usertable",{_id:"hashed"},false,{numInitialChunks:200})

   END_CMD

3. Use workload_s1 and cluster instances as an example. Run the following command to prepare data:

   ./bin/ycsb load mongodb -s -P workloads/workload_s1 -p mongodb.url=mongodb://rwuser:${password}@${mongosIP}:8635,${mongosIP}:8635/ycsb?authSource=admin -threads ${threadNum} 1>workload_s1_load.result 2> workload_s1_load.log

4. Use workload_s1 and cluster instances as an example. Run the following command to test the performance:

   ./bin/ycsb run mongodb -s -P workloads/workload_s1 -p mongodb.url=mongodb://rwuser:${password}@${mongosIP}:8635,${mongosIP}:8635/ycsb?authSource=admin -threads ${threadNum} -p maxexecutiontime=1800 1>workload_s1_run.result 2> workload_s1_run.log

   

   • ${mongosIP} indicates the private IP address of the mongos node in the DDS cluster instance.
   • ${password} indicates the administrator password of the DDS instance.
   • ${threadNum} indicates the number of concurrent threads for running the test. In this test, the number of concurrent threads is 128.

Testing Models

• Workload model

  Table 3 Service model

  Service Model No.	Service Model
  S1	100% insert
  S2	90% update ,10% read
  S3	65% read ,25% insert, 10% update
  S4	90% read ,5% insert, 5% update
  S5	50% update, 50% read
  S6	100% read

• Number of concurrent threads: 128

• Document model

  Use the default configuration of YCSB: The size of each document is 1 KB, and the default index is _id.

• Data volume to be prepared

  In this test, prepare two types of data volume for each cluster instance.

  For details, see the following table.

  Table 4 Data volume to be prepared

  Specifications	Low-Level Data Volume	High-Level Data Volume
  2 vCPUs | 4 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  2 vCPUs | 8 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  4 vCPUs | 8 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  4 vCPUs | 16 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  8 vCPUs | 16 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  8 vCPUs | 32 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  16 vCPUs | 32 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  16 vCPUs | 64 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  32 vCPUs | 128 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000
  64 vCPUs | 256 GB	Storage space: 10 GB Record counts: 10,000,000	Storage space: 100 GB Record counts: 100,000,000

• Data consistency model

  Weak consistency: For the write concern setting of {w: 1, j: false}, an acknowledgment is returned after data is written to the disk on a single node. Data is persisted on disks in asynchronous mode with the default write concern setting.