apache
Update ProfilingApplication default options to align with HTTP.
Some performance numbers to compare the current master-http branch against 3.7-dev using ProfilingApplication.
ProfilingApplication Information
The ProfilingApplication has two modes: latency and throughput.
The latency test will run a particular query one at a time for several runs. It then reports the average latency of the query. The chosen query for this test was g.V().both().repeat() as the server is able to quickly respond to this request but it also generates a lot of data to be returned over the wire.
The throughput test will issue requests (an application argument) number of queries to the server and see how long it takes to complete them. It will then report the results as number or queries completed per second. The throughput tests can be either exercise or 1+1. When the application argument exercise is set to true, there are a number of preset queries that will be used, if it is false, then "1+1" is the query used. See https://github.com/apache/tinkerpop/blob/master/gremlin-driver/src/main/java/org/apache/tinkerpop/gremlin/driver/util/ProfilingApplication.java#L54 for the list of queries
Performance Metrics
Test System
Server EC2 m7a.8xlarge US EAST 2
Server Config
maxWorkQueueSize: 65536
threadPoolWorker: 8
gremlinPool: 16
evaluationTimeout: 30000000
Driver EC2 m7a.4xlarge US WEST 2
| Branch | Test Type | Arguments | Query | Result |
|---|---|---|---|---|
| 3.7-dev | Latency | minConnectionPoolSize 1 | g.V() 6 vertices |
0.05625982837 sec |
| 3.7-dev | Latency | minConnectionPoolSize 1 | g.V().repeat(both()).times(15) 2 824 992 vertices |
3.844317 sec |
| 3.7-dev | Latency | minConnectionPoolSize 1 | g.V().repeat(both()).times(18) 39 750 654 vertices |
53.4815487 sec |
| master-http | Latency | minConnectionPoolSize 1 | g.V() 6 vertices |
0.05744388 sec |
| master-http | Latency | minConnectionPoolSize 1 | g.V().repeat(both()).times(15) 2 824 992 vertices |
3.915586 sec |
| master-http | Latency | minConnectionPoolSize 1 | g.V().repeat(both()).times(18) 39 750 654 vertices |
54.892998565 sec |
| 3.7-dev | Throughput | parallelism 16 requests 10000000 | 1 + 1 |
35458 req/sec |
| 3.7-dev | Throughput | parallelism 16 requests 10000000 exercise true | N/A |
30159 req/sec |
| 3.7-dev | Throughput | parallelism 16 requests 100000000 exercise true | N/A |
30488 req/sec |
| master-http | Throughput | parallelism 16 requests 10000000 | 1 + 1 |
19758 req/sec NOTE: the pool size is small so the performance is lowered |
| master-http | Throughput | parallelism 16 requests 10000000 exercise true | N/A |
19690 req/sec NOTE: the pool size is small so the performance is lowered |
| master-http | Throughput | parallelism 16 requests 10000000 minConnectionPoolSize 1500 maxConnectionPoolSize 5000 | 1 + 1 |
34450 req/sec |
| master-http | Throughput | parallelism 16 requests 10000000 minConnectionPoolSize 1500 maxConnectionPoolSize 5000 exercise true | N/A |
32138 req/sec |
| master-http | Throughput | parallelism 16 requests 100000000 maxConnectionPoolSize 5000 exercise true | N/A |
30063 req/sec |
