Consensus of the Benchmark Test

The latest article about quorum performance test I can find was written in 2018. This is interesting, as one of the most widely used blockchain system, can you imagine that there isn't any article about it's performance? So I did the performance test myself.

Hardware:

5*8C16G50GSSD as quorum servers

5*4C8G20GSSD as clients to send requests

Software:

System: CentOS 7

Quorum: 1.10.2-stable (quorum-v22.4.4)

Java:OpenJDK 1.8

Mysql:5.7

Quorum-Benchmark: A benchmark program written in Java, developed by myself.

Test step:

  1. Create 200,000 Accounts
  2. Send ether to all accounts above
  3. Using accounts above to sign raw transactions, store the hexed string result in file
  4. Back up the blockchain using ‘geth export’ command
  5. Send raw transactions stored in file to the quorum network
  6. Statistics the result
  7. Repeat the 5th step, using different request rate

Test Result:

about 1491TPS

about 1347TPS

Conclusion

As the picture above, you can see the blockchain system appeared unstable when the tps reached about 1350+, the block interval is more than 5s as we set, and there were lots of empty blocks.

From my test result, I got the best TPS is about 1250 which the system can still keep stable.The cpu and memory were both low, so I think the bottleneck is the QBFT consensus.

Guess you like

Origin blog.csdn.net/HardRedStone/article/details/125935987