@lukastribus I always thank you for your help.
I understand checking the memory and cpu after measuring how much traffic are there and the number of concurrent connections.
And I’ll read carefully the documents you said, for pinning a haproxy to cpu.
I have clues about what I should do.
Thanks you so much. 
This very insightful advice from Lukas is the most important rule to always keep in mind. Indeed, most bottlenecks found in field that are hard to work around are caused by unneeded over-engineering. The right way to optimise is by trying to reduce the number of machines once you have something that works well. In this case you’ll have solid numbers to discuss about and you’ll have references to compare the effects of certain actions.
1 Like
@willy I really thank you for your advice.
I keep your words in my mind. I think I can do better in the future thanks to you. 