How to increase the node qps for both jsonrpc and http api?

[PikaZ76]

I'd like to deploy a Tron full node used to sync data through it's jsonrpc and http api, is there any solution in configuration to hit the limition of the server?
I got a 64 core CPU, 256 GB Memory.

and I used this configuration:

node {
  tcpNettyWorkThreadNum = 128

  udpNettyWorkThreadNum = 8 
}

rpc {
    thread = 32
}

rate.limiter{
  # global qps, default 50000
  global.qps = 500000
  # IP-based global qps, default 10000
  global.ip.qps = 500000
}

so, are there any way to do in configuration file or JVM to maxmum the api performance?

[angrynurd]

@PikaZ76
Based on your machine specifications (which are quite high-end), I would recommend the following configurations to maximize resource utilization:

1. Increase JVM memory allocation in startup command:

# Example: increase to 128GB
JAVA_OPTS="-Xms128g -Xmx128g"

2. Increase QPS limits:

rate.limiter = {
  global.qps = 500000
  global.ip.qps = 500000
}

3.HTTP configuration:

node {
  maxHttpConnectNumber = 500  # increase from default 50
}

When adjusting these configurations, please tune them gradually based on monitoring metrics:

If CPU usage is too high, consider reducing thread count
If memory usage is too high, reduce maxHttpConnectNumber
If response time increases, lower QPS limits

[PikaZ76]

any solution on leveldb config?

[angrynurd]

any solution on leveldb config?

@PikaZ76
If you are using a mainstream SSD drive, the query latency of LevelDB is very low (compared to the overall latency of JSON-RPC/HTTP interfaces). Typically, in a single API query, LevelDB only takes about 5-10ms, which is already very fast. The disk(especially SSD drive ) is usually not the bottleneck.

If you still want to improve LevelDB performance, you can try modifying the following parameters in the storage section of the main_net_config.conf file:

1.cacheSize: Recommended to set a larger value (like 4GB or more) to improve cache hit rate
2.maxOpenFiles: Set a larger value to avoid frequent file open/close operations