We found funny specificity when using x86_64 on server platform.
Example. We have server Dell R720 with next hardware setup (one of typical NATs per 2k users):
- 2 CPUs x Xeon E5-2650L v2
- basic clock 1,70 GHz
- turbo boost 2.0 up to 2,10 GHz
- 10 cores / 20 threads per CPU
- 25 MB smart cache per CPU
- energy efficient 70 W
- 16 GB RAM DDR3 ECC
- network X520-DA2
- run from USB
On other side - test setup on old desktop hardware:
- Intel i5-3330
- 4 cores / 4 threads
- basic clock 3,00 GHz
- boost clock 3.20 GHz
- 6 MB Smart Cache
- 6 GB RAM
With default setup settings, old desktop is faster (latency, single client performance) on low-load, because have higher CPU clock.
Solution:
1) disabling Hyper-Threading / leaving only logical CPUs
2) disablin C-States and Energy-saving
3) enabling Performance Profile or custom high performance with no clock downgrade
For today ROS <6.48.6 (long-term) dont have multicore optimization.
Hiya.. we have some old R910 and R920.. 4 CPUs 10cores/20threads each cpu clock i think 2.6ghz or 2.8ghz will have to check.. i need to setup a CGNAT server for 2K with expansion up to 5k.. would i be better of running long-term v6.48.xx in bare metal no vmware wise.. or RoS v7? intel x520 dae is fully compatible with v6 routerOS? as we have tried some HP NCxxxx dual slot cards.. and it did not recognize neither in v6 or v7 beta