Scimore TPC-C test
To get an idea of the performance improvement for version 2, we implemented the TPC test for ScimoreDB. We wanted to test the performance and durability of our innovative scale out solution. The Technical University of Denmark (DTU) provided machines for the tests.
We tested our tpc-c implementation on clusters of 1, 3, 7, 15 and 31 machines.
The physical setup
The machines in the test where 2.13 Ghz Duo Core, 2Gb memory, single sata disc standard
Dell office machines for a price of approximately 1.000$. For the cluster configuration
with 1, 3 and 31 machines, the database was configured to use 150.000 pages of memory
pr. machines, corresponding to 1.14Gb. For the clusters with 7 and 15 machines,
each database was configured to use 110.000 pages, corresponding to 839Mb memory.
All machines where connected using a gigabit network.
Results
The results shows that ScimoreDB scales in a distributed environment.
| Machines | inserts/s | updates/s | selects/s | queries/s | tpc-c trans/s |
| 1 | 3000 | 3000 | 2500 | 8500 | 530 |
| 3 | 7300 | 7300 | 5300 | 19900 | 1150 |
| 7 | 15000 | 15000 | 13000 | 43000 | 2500 |
| 15 | 29000 | 29000 | 25000 | 83000 | 5000 |
| 31 | 59000 | 59000 | 49000 | 167000 | 9200 |
First a graph showing the average number of sql queries pr. second on the database
cluster.
Next a graph to illustrate the type of queries executed on the cluster simultaniously
during the test. This shows on average how many inserts, updates and selects are
executed pr. second.
The precautions: This not a certified tpc benchmark. We did implement the standard tpc benchmark. The test was performed without the 2 hours warmup, instead we did a touch of data on all tables to ensure sufficient data was in memory. The resulting queries/s numbers where extracted for the system performance tables of scimoredb.
Comments
We are very greatfull for DTU providing hardware and assistance with the trial setup.
Thanks!