I had router on Pentium 4, simple queues+mangle utilize about 90% of CPU;
Than i upgrade it to Core 2, with 2 cores, than traffic is smaller than CPU usage is 30% if traffic is larger CPU usage is 50%, and cant go bigger (queues+mangle cant utilize more than one core?).
If i switch multi-cpu=no than respectively 60% and 100%.
Then i tested on Quad core, CPU usage always 25% (also queues+mangle can utilize only 1 core?).
Version 3.28, but this ploblem was on all 3.x.

So my problem is that i cant search CPU that will hold my load. As working only 1 core, now i plugged Core 2 E8600 3.33 Ghz, now my max load is 45%, but i worry than after few month my load will be greater, and problem comes back.

cluster ruleZ =)

I also ran into this error

I also encountered this error


The network card are more. The mistake is more serious


I can not speak English.

3.x has this bug

actually, it's not a bug. it's a feature =) or feature absence

linux has this problem itself ->all linux kernel

even super-puper-mega-kernel cannot parallelize single thread

O.k i understand, second question:
Is mangle, firewall, queues, packet bridging, and passing througt PC from one interface to another, maybe routing too - one thread?

So as i understand second core may be used just for threads like DHCP server or NTP synchronizing? LoL

And what i can do if the fastest Core 2 processor can't shape my 100 Mbps flow with one core?

But i plan to increase the flow at least to 200 Mbps..

it is possible, but it has to be quite optimised setup.

for example - PCQ instead of "1 simple queue per customer" etc.

Janis, isn't it planned to make multi-cpu optimization of routing, etc? in v4, for example =)

When can this question solve?

600Mbps had problems time out..

CPU use 33%

Perhaps we can hope for linux integration with 'Grand Central Dispatch' now that it's become open source.
This could solve some problems with multi-core systems and give Mikrotik some new options to look at

Link: http://developers.slashdot.org/story/09 ... h?from=rss

There are a lot of ways but the only problem is LAZY programmers
CUDA Programming

Any news in running MUM about this topic ? seems v5 will available in two weeks ?