I use the processor with 4 kernels, at loading in 300 pptp sessions loading CPU of 25 percent, at 500-600 sessions loading CPU same. Why MT uses only one kernel!!! When it will be corrected to work not probably!!!
olmi:
You can try using “virtual load balancing”. Use kvm to create four virtual RouterOS machines where each terminates your tunnels. Then create firewall nat rule which will load balance connections between virtual machines. That way You will be able to use all four cores because each core will run separate RouterOS instance. Overhead of using virtualization shouldn’t be that large.
I have to agree about the propeller unfortunately…
Virtualization is a great option, but can I avoid the freezes that I was seeing once the
load goes over 500 people (versions 3.23 up to 4.4). Sending a bunch of files to support
produced no results. They suggested to install a debug package, which wasn’t quite what we wanted.
It’s difficult to blame the hardware because we’ve tried about five different systems with
different configurations from Xeon on Supermicro to P4 with 3GB on asus.
As for the processor load on one of the cores,
we have a system (Quad Core ~ 2. with a good example:
It’s running NAT, ~40 queue tree items. The traffic is ~220-250 Mbit/s, over 50 kpps.
As soon as we hit the performance limit of one core we get a drop in traffic on this router.
I’m not even talking about quality of service or high speeds in clients’ tests.
How can this situation be resolved without switching to another platform?
When can we expect version 5 to come out?
is your software not meant for high loads? many have paid extra money to keep working on your platform!
If it is started a little pppoe servers on different VLAN is will be considered as different processes?? And whether it will be scattered processes on kernels?
with a good example: