if my vm host has 64 cores, and i give 8 cores to 10 vms that are running simultaneously (obviously assigning more cores than physically exist) then what happens? does bhyve refuse to start vms when they've all been assigned or do they just start sharing what capacity does exist?

I would imagine you'd have CPU contention and they'd start behaving just as behave other processes under CPU contention

efficiency would go down due to cpu time being wasted on contention battles i guess?

so i have windows 11 iot guest vm installed on fresh freebsd 14.4 + bhyve-vm. i gave it 16 cpu cores as guides say to: cpu=16, cpu_sockets=1, cpu_cores=16, cpu_threads=1, and it's very very sluggish. even idle with nothing running it uses 20-50% of the cpu. on the host side, running sudo top, i see it using 800% WCPU, and keeping host at 85% idle

or so (dell r430 with 64 cores across dual xeon) anyone have any clue?

i mean i know it's windows but still, this doesn't seem right

kerneldove: If they're all running at 100% at the time then you would have issues.

Over subscribing CPUs in a vm env is pretty common only because you're going to 100% all the time.

Polarity?

skered you mean NOT going to use 100% all the time?

yea not

skered is it safe to oversubscribe cores as long as idle never reaches a sustained 0%?

I'm surprised that #FreeBSD is almost 30% of #Linux's size here on LiberaChat.

you think we should kick 350 people to balance out the size?

Only if you're kicking them for #linux, but but I think you'd need to kick a lot more than 350.

s/for/from

If someone knows how to convince benchmarks/kdiskmark to work I would love to hear it. All I get from it is "Benchmark Failed / No such file or directory" when pressing the button to test.

fwiw I am running it with `sudo kdiskmark` to sidestep the need for permissions escalation.

linuxer:

joepublic: it seems its icons its missing /usr/local/share/icons

and a ton of them

Interesting. I'll have to chase that down and devise a solution I guess.

Thanks.

though i doubt that's it, truss -o check.log kdiskmark

see what it wants to open in check.log

output a bunch of .. lets call it 'things'

there are probably better apps for, i still typically just dd or bonnie++ if i feel like staring at more numbers

after extensive testing for always on vpn on freebsd

I have concluded it will need coding :p

the best I got is all traffic but local traffic via wg

the issue is the routing table will always route local traffic first as /64 is smaller than /0, and /24 is smaller than /0

without said rules though, the phy cant route to the router, and static addressing isnt an option if you are using lots of public networks

a kernel option would likely be needed which would force all traffic via wireguard if enabled

if its not the wireguard endpoint, then ignore the routing table and go via wg

idk :p

Android has it implemented well

I need to work on captive portal detection script

detect_204, if you will

iRobbery: I need to bail, but I want to revisit your jail/nat stuff sometime. I've set up similar here and it's deeply pleasing. Switching from ipfw to pf was critical to it being simple and happy.

polarian: how about using fibs for that? I think creating your wg interface with fib 1 tunnelfib 0 having your normal routing on fib 0 and set your default fib to 1 should work

I havent played with fibs :)

in fact I know nothing about it

u got some docs for it

ahhh

that would work well

just read a little bit of the setfib manual page

and then processes which need to bypass the vpn (like a captive portal check) can be switched to fib0

polarian: Random thing I did a while back - needed different default routes for different services: bpa.st/WKLA

bbl

will look into it thx