Difference in PTR and PROD cpu-usage
-
Hey all!
While I was testing stuff in PTR I found out that PTR uses about 30% less CPU. Is there any reason the PTR reports less cpu usage?
I've got the exact same setup and exact same amount of creeps on PTR compared to PROD, yet prod is chugging along at ~180ms and PTR is arround 125.
Any reason for such a big difference?
-
I can confirm this on my end as well. Code on PTR consistently runs at about 30% less CPU than on the main server (or to put it another way the main server takes 50% more CPU than the PTR server does for the same code).
-
<p>Since PTR is on new branch with the new flags format, it could be related to some optimizations we've made here. Let's wait till May 12 with this.</p>
-
Oh cool, that would explain the difference between me (427 flags) and tedivm (855 flags) findings
Will we get a detailed change report? I'm curious to what is changed that the flags could be so much faster.
-
Since the update the CPU has roughly stayed the same on PROD, the code on PTR still seems be a lot faster. This makes it quite difficult to create new features and push them to prod, as they might impact cpu much more on PROD than they do oin PTR
What is causing this huge difference?
-
Well, frankly, it is hard to reason anything here. The equipment is exactly the same, just uses only 1 core instead of 16. There are probably some indirect factors like garbage collection being triggered only for a few users, and thus having less impact on every single user, but it is very hard to measure and eliminate.
-
Ahww that's a bummer. I wish I could find out what's causing the difference.
