blockworks-foundation
/
solana-rpc-testing
mirror of https://github.com/blockworks-foundation/solana-rpc-testing.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

update bench with one CPU for Actix

This commit is contained in:
musitdev 2023-07-14 13:26:28 +02:00
parent b8269e8f34
commit 2826fe2ac7
1 changed files with 58 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
bombardier_tests.md
View File

@ -132,6 +132,8 @@ Time out limit: 2500
### jsonrpc_actix branch
#### Using default worker model (use all CPU - 5)
125 concurrent connections
```
@ -180,6 +182,59 @@ Statistics Avg Stdev Max
Time out limit: No time out at 10 000
#### Using 1 worker 1 thread
125 concurrent connections
```
~/go/bin/bombardier -m POST http://192.168.88.12:8899 -c 125 -n 1000000 -l -H "Content-Type: application/json" -b '{"jsonrpc": "2.0","id": 1,"method": "getBalance","params": ["4EVAJ81v9fLjz2wp44SZmxgTbvYB8dBRZ6s1SAqP99eZ"]}'
Bombarding http://192.168.88.12:8899 with 1000000 request(s) using 125 connection(s)
1000000 / 1000000 [=====================================================================================================================================================================] 100.00% 70352/s 14s
Done!
Statistics Avg Stdev Max
Reqs/sec 71166.10 3382.88 78372.49
Latency 1.75ms 72.38us 7.97ms
Latency Distribution
50% 1.74ms
75% 1.76ms
90% 1.79ms
95% 1.86ms
99% 2.00ms
HTTP codes:
1xx - 0, 2xx - 1000000, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 30.13MB/s
```
#### Using 1 worker 512 threads
125 concurrent connections
```
│~ via 🐏 17GiB/31GiB | 0B/33GiB
└─> ~/go/bin/bombardier -m POST http://192.168.88.12:8899 -c 125 -n 1000000 -l -H "Content-Type: application/json" -b '{"jsonrpc": "2.0","id": 1,"method": "getBalance","params": ["4EVAJ81v9fLjz2wp44SZmxgTbvYB8dBRZ6s1SAqP99eZ"]}'
Bombarding http://192.168.88.12:8899 with 1000000 request(s) using 125 connection(s)
1000000 / 1000000 [=====================================================================================================================================================================] 100.00% 72391/s 13s
Done!
Statistics Avg Stdev Max
Reqs/sec 72981.41 3820.08 78156.79
Latency 1.71ms 82.02us 9.94ms
Latency Distribution
50% 1.70ms
75% 1.72ms
90% 1.74ms
95% 1.78ms
99% 1.93ms
HTTP codes:
1xx - 0, 2xx - 1000000, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 30.90MB/s
```
### Worker model conclusion
The main diff between jsonrpsee execution and Actix seem to come from the CPU use. Jsonrpsee use on CPU whereas the default worker model of Actixw use all CPU.
With one CPU Actix perform a little a little worse than jsonrpsee.
## Test2: getBlocks
@ -504,17 +559,9 @@ Statistics Avg Stdev Max
The tokio_jsonrpsee show a little improvement for both request compared to master branch.
For the getBlocks the jsonrpc_actix show an important improvement of the performance. I've to find the reason because I don't think it's only related to the json serialisation and HTTP request handling. This improvement mus tbe confirmed with the getBalance request when it will be available.
Actix show an important improvement of the performance. It seems to come from a better CPU usage of Actix. As there's no other processing on the validator, all CPU are available.
## Actions
Forcing Actix with one CPU, is perform a little worse than jsonrpsee.
- correct getBalance on jsonrpc_actix
- Do the getBalance test for jsonrpc_actix branch
- See the reason of the performance difference between jsonrpc_actix branch and the others.
So it show that the execution model is important for the overall performance and with a 1 CPU execution, jsonrpsee perform a little better.
GetBLockHash GetSlot.
curl http://localhost:8899 -X POST -H "Content-Type: application/json" -d '
{"jsonrpc":"2.0","id":1, "method":"getSlot"}
'