Community discussions

MikroTik App
 
hooyao
newbie
Topic Author
Posts: 37
Joined: Mon Feb 20, 2017 6:11 pm

v7 is significantly slower, at least on x86

Sat Sep 24, 2022 5:59 am

I intended to replace my RB4011 with x86 routers, it's little bit tretched when handling 1000M download. 5009 is way too expensive and out of stock most of the time. I started experimenting with the N3160/dual RTL8111H eth.

I installed v7.5 with default router config w/wo fasttrack as the 2nd level router in my home LAN, the result was disappointing. Openspeedtest & iperf in my home LAN can only reach 600Mbps/600Mbps download/upload, cpu reaches 80% for one core.

Then I downgraded it to v6.49.6, with the same configuration, the download/upload speed can easily hit 940Mbps, cpu reaches 80% for one core.

Is there tweaks I can do to boost v7 performance on x86? Or Do I have to get a much more powerful CPU to run v7?
 
sindy
Forum Guru
Forum Guru
Posts: 10205
Joined: Mon Dec 04, 2017 9:19 pm

Re: v7 is significantly slower, at least on x86

Sat Sep 24, 2022 9:37 am

First, the linux kernel under v7 doesn't support route caching, so v7 will be slower in most scenarios.
Second, CHR doesn't actually fasttrack traffic (although it indicates the connections to be fasttracked) even on v6, I'd assume x86 to be the same case.
Third, when you say "CPU reaches 80% of one core", do you mean that the other cores are loaded much less? If so, do you test the bandwidth using a single TCP (UDP) connection or multiple ones?
 
hooyao
newbie
Topic Author
Posts: 37
Joined: Mon Feb 20, 2017 6:11 pm

Re: v7 is significantly slower, at least on x86

Sat Sep 24, 2022 6:10 pm

First, the linux kernel under v7 doesn't support route caching, so v7 will be slower in most scenarios.
Second, CHR doesn't actually fasttrack traffic (although it indicates the connections to be fasttracked) even on v6, I'd assume x86 to be the same case.
Third, when you say "CPU reaches 80% of one core", do you mean that the other cores are loaded much less? If so, do you test the bandwidth using a single TCP (UDP) connection or multiple ones?
I did the bandwidth test with openspeedtest and iperf -P 4, i guess they establish multiple connection to the server. n3160 has 4 cores, the busiest core could reach 80%, others are 20% to 60%.
I guess the linux scheduler dispatches the load to all the cores instead of affiliating it to a specific core. The max CPU usage was 80%, it seems v7 won't juice the last bit of CPU.
As for route cache, as a home router, downloading large amount of data via several connections is the primary use case. It is disappointing.

Who is online

Users browsing this forum: CGGXANNX, seriosha and 81 guests