This document describes how to perform standard performance testing on TencentDB for MongoDB instances. The resulting performance data is for your reference only.

Test Environment

• Test date: August 2020
• Client specification: the client is installed on an 8-core CVM instance with 32 GB memory. Tests show that if the MongoDB instance has a low specification, the CPU utilization of its replica instance can reach 100% under the request pressure of an 8-core CVM instance with 32 GB memory. In this case, one CVM instance achieves even better test results than multiple CVM instances can do. However, if the CPU utilization cannot reach 100% when you use only one CVM instance, you can try 4 CVM instances to share the concurrent threads.
• Test object: TencentDB for MongoDB v4.0 replica set instance

Testing Tool

Download YCSB.

Test Overview

Prepare about 10 GB data. Then, run 100 or 200 concurrent threads and use YCSB to test the throughput (ops/sec), RAL (read average latency in us), and WAL (write average latency in us) of instances with different specifications under the pressure of 50% read requests and 50% update requests, or 95% read requests and 5% update requests.

Latency

The average latency from CVM to MongoDB is 0.35 ms.
Minimum latency = 0.30 ms, maximum latency = 0.44 ms, average latency = 0.35 ms

Commands used in the test

1. Prepare data (about 10 GB)

nohup ./ycsb-0.15.0/bin/ycsb load mongodb -s -P workloads/workloada
-p mongodb.url=mongodb://mongouser:password@10.xx.xx.30:27017,10.xx.xx.28:27017,10.xx.xx.5:27017/admin?w=0 -p table=test -threads 300 -p recordcount=10000000>loadlog.txt &

2. Test under 50% read requests and 50% update requests

nohup ./ycsb-0.15.0/bin/ycsb run mongodb -s -P workloads/workloada -p mongodb.url=mongodb://mongouser: password @10.xx.xx.30:27017,10.xx.xx.28:27017,10.xx.xx.5:27017/admin?w=0 -p table=test -p recordcount=10000000 -p readproportion=0.5 -p updateproportion=0.5 -p insertproportion=0 -p operationcount=100000 -threads 100 >runlog.txt &

3. Test under 95% read requests and 5% update requests

nohup ./ycsb-0.15.0/bin/ycsb run mongodb -s -P workloads/workloada -p mongodb.url=mongodb://mongouser: password @10.xx.xx.30:27017,10.xx.xx.28:27017,10.xx.xx.5:27017/admin?w=0 -p table=test -p recordcount=10000000 -p readproportion=0.5 -p updateproportion=0.5 -p insertproportion=0 -p operationcount=100000 -threads 100 >runlog.txt &

Note：

• Modify -p operationcount=100000 according to the specific execution time to make sure that the execution time is longer than 20 minutes; otherwise, the resulting data will not be representative.
• w in ?w=0 represents write concern.
  • w:1 (acknowledged write) requests acknowledgment that the write operation has propagated to the standalone mongod or the primary in a replica set. w:1 is the default write concern for MongoDB.
  • w:0 (unacknowledged write) requests no acknowledgment of the write operation. w:0 will not report on the success of a write operation, but may return information about socket exceptions or networking errors when data is written to a closed socket or network is unavailable.
  • w:>1 (valid in a replica set) requires acknowledgment from the primary and as many data-bearing secondaries as needed to meet the specified write concern.

Test Data

Read/update ratio (50:50)

Read/update ratio (95:5)