TNSR CPU utilization on SG-5100
-
I am in the process of setting up the latest TNSR version on an SG-5100. Am in the process of learning the ins and outs of TNSR so the configuration is very basic right now. I have one interface dedicated to the OS, and then the rest of the interfaces are allotted to TNSR. Aside form the WAN interface, I have configured a primary LAN (10.0.2/0/24) on one interface and a secondary LAN (10.0.20.0/24) on another interface. A couple simple NAT rules, DHCP on both interfaces, and that's about it. Just ran top and noticed that vpp_main is utlizing ~25% of my CPU potential (i.e. 1 of 4 cores). Is this normal behavior? I don't think I've done enough to screw something up yet. :) Thanks!
-
That is completely normal. The dataplane will always consume 100% of one CPU core
-
Outstanding! Thanks for the reassurance that I haven't bjorked something already!
-
Follow on question - is the dataplane activity always using 100% of the core (regardless of load) or is the core merely 100% reserved for the dataplane and utilized as needed? Hope that makes sense. If the former, would that cause any degradation of the core over time after constantly being hit at 100%?
-
It's the former, and as far as I'm aware there shouldn't be any detrimental effects other than possibly heat. But I run it on a couple SG-5100s for testing and they barely get warm to the touch. I don't push them very hard, though.
-
Thanks as always for the response @jimp. I wasn’t overly concerned. Just curious how it all works.
-
It's not 100% the same but think kind of like the old polling option in pfSense. Rather than wait for hardware interrupts, it's always running a tight loop looking for data to process.
-
That makes a lot of sense. A system, or person for that matter, waiting to be told to work will never be as responsive as one that is at the ready to pounce.
-
G gabacho4 referenced this topic on
