Hyperlambda is 7 times Faster than N8N
I wanted to measure the performance of N8N versus Hyperlambda, so I used the Docker version of N8N, and I created a simple workflow that returns all rows from my chinook SQLite database and its Artist table.
Notice, you have to use the "Command" node and execute SQLite as a command to make this work in N8N, and make sure your first node is a webhook that returns the response from your last node.
Then I used a Python script that creates 50 worker threads, each thread "hammering" my HTTP endpoints, counting how many invocations it can execute in 30 seconds.
- Hyperlambda 14,500 HTTP requests
- N8N 2,150 HTTP requests
The time period was the same, 30 seconds, and both examples were using Docker, since that's the preferred way to install N8N locally. The end result implies Hyperlambda can deal with 7 times as much throughput as the equivalent N8N solution. With other words, if your N8N solution scales to 2,000 concurrent users max, the Hyperlambda equivalent would scale to 14,000.
The N8N workflow was a fairly simple workflow. It's just a simple webhook returning the result of the last node, with a "command" node that connects to SQLite and returns rows for me as JSON.
Hyperlambda performance
Although N8N is far superior to LangChain, and JavaScript performs much better than Python, it still suffers from some of the same problems as Python and LangChain.
Hyperlambda is just a tiny abstraction on top of C#, while N8N is a visual workflow designer. It's created very well, and it was easy to get started with it - But such "visual workflows" will never be able to perform at the same speed as a "true programming language". The problem is that the composable "nodes" aren't executing at anything even resembling the speed of a real programming language. Basically, the more "logic" you add to your workflows, the slower they become. Hence, I suspect that in a real world application, with workflows containing potentially dozens of nodes, maybe hundreds, the difference in performance would increase significantly. This was just a tiny workflow with two nodes though, so the difference was "only" 7 times.
Conclusion
This implies that if you were using Hyperlambda instead of N8N, you'd have at least 7 times the performance, probably more for complex workflows with a lot of nodes.
