This is a benchmarking of popular databases to get a view of the vanilla write and read TPS that is possible against these databases.
NodeJS is used to expose a write and read API for some of the popular datastores. Each individual folder has basic documentation explaining how the test can be run for each datastore.
v1/addUser POST API is used to benchmark the writes with the following payload -
{
"id": ${counter},
"userName": "testuser${counter}",
"userDetails": "A really long list of user details. This is just a random string to ensure test data in the column."
}
/v1/readUser/:id GET API is used to benchmark the reads
- The perftest folder contains a jmx file that has the above mentioned APIs.
- The database & the nodejs APIs were run as docker containers. The database was limited to 3 cores and the API was limited to 5 cores as can be seen in the individual documentation.
- Both the containers were running on the same machine along with the Jmeter (ver 5.3) instance.
- Machine had 6 cores (Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz) with 2 threads per core and 32 GB of RAM
- None of the databases/API containers was using more than 4 GB of RAM during the test although there were no limits applied
The test observations are the TPS/avg response time while writing 20 million records to the databases. The reads were done against the 20 million records that were added during the write benchmarking
| Database | Operation | Request Per Second | Avg Reponse Time (millisecond) | Observations |
|---|---|---|---|---|
| Postgres 12 | Write | 7500 TPS | 13 | NodeJS and DB utilizing all allocated cores |
| Postgres 12 | Read | 10500 TPS | 9 | NodeJS and DB utilizing all allocated cores |
| ElasticSearch 7.8 | Write | 6000 TPS | 16 | NodeJS and DB utilizing all allocated cores |
| ElasticSearch 7.8 | Read (Get by document ID) | 12000 TPS | 8 | NodeJS and DB utilizing all allocated cores |
| Cassandra 3.11.6 | Write | 14000 TPS | 7 | NodeJS and DB utilizing all allocated cores |
| Cassandra 3.11.6 | Read | 10500 TPS | 9 | NodeJS and DB utilizing all allocated cores |
| Scylla 4.0 | Write | 15500 TPS | 5 | NodeJS and DB utilizing all allocated cores |
| Scylla 4.0 | Read | 18000 TPS | 5 | NodeJS and DB utilizing all allocated cores |
| Redis 6.0 | Write | 24000 TPS | 4 | NodeJS utilizing all allocated cores, Redis was using .5 core of 3 allocated |
| Redis 6.0 | Read | 28000 TPS | 2 | NodeJS utilizing all allocated cores, Redis was using .5 core of 3 allocated |