TheJollyRoger has quit [Remote host closed the connection]
TheJollyRoger has joined #spectrum
FireFly has joined #spectrum
multi has quit [Quit: beep?]
multi has joined #spectrum
lukegb has quit [Read error: Connection reset by peer]
lukegb has joined #spectrum
IdleBot_1c746da8 has quit [Ping timeout: 260 seconds]
puckipedia has joined #spectrum
qyliss- has joined #spectrum
Yakulu[m] has quit [*.net *.split]
danielrf[m] has quit [*.net *.split]
M011100110110110 has quit [*.net *.split]
kenmacd[m] has quit [*.net *.split]
rhysmd has quit [*.net *.split]
puck has quit [*.net *.split]
qyliss has quit [*.net *.split]
puckipedia is now known as puck
qyliss- is now known as qyliss
Yakulu[m] has joined #spectrum
M011100110110110 has joined #spectrum
kenmacd[m] has joined #spectrum
rhysmd has joined #spectrum
danielrf[m] has joined #spectrum
moonloo has joined #spectrum
<moonloo>
is there even a need for anything custom made? i have been unable to notice any meaningful latency with just waypipe over vsock, and by adding nothing no attack surface is added.
<moonloo>
you can use multiple network namespaces on the host and through socat's ability to create TUN interfaces network your guests as members of multiple network namespaces on the host. the problem of intervm communication is thus trivially solved unless you have some really unusual use cases that would bottleneck at the vsock or network stack interface.
<mvnetbiz_>
I think some people are more sensitive to ui latency
<moonloo>
mvnetbiz_ : so i just tried some browser games on firefox over waypipe running on a guest, also mouseaccuracy.com and as not able to notice any latency. and you arent going to be gaming on vms that have no gpu passthrough anyway. im sure there are a lot of "needless" buffer copies going on but they are all local. and for that price you are geting 0
<moonloo>
additional attack surface and minimal effort to set up required.
<colemickens>
virtio-fs > 9p btw
<colemickens>
I haven't seen anyone actually use it in nix land yet, but we should and someone opened an issue about it (it might help some mac use cases IIRC?)
<moonloo>
9p seems to also be usable over unix sockets? "unix - specifying a named pipe mount point" with some socats it can be made to go over vsock. does ti compare favorable in performance to NFS I have been unable to find anything on that front. I know that NFS over vsock seems to be dead, and it's not even usable via sockets - TCP seems like
<moonloo>
unneeded overhead for VMs.
moonloo has quit [Quit: Ping timeout (120 seconds)]
moonloo has joined #spectrum
<mvnetbiz_>
moonloo: what are you using again? Waypipe and socat with software rendering on the application? Or do you mean just not whole pcie card gpu passthough
<mvnetbiz_>
Nvm I found your explanation above, but I'm curious about the software rendering
<moonloo>
there is no dri in the vm. waypipe gives the guest a wayland-0 sockets and apps use it, so software rendering only.
<moonloo>
in my setup is one app per vm though, firecracker spawns them in like 200ms. init takes longer but i haven't optimized any of that. i plan to just use alpine and open a container from NFS or 9p over vsock.
moonloo has quit [Quit: Ping timeout (120 seconds)]
cole-h has quit [Quit: Goodbye]
cole-h has joined #spectrum
mvnetbiz_ has quit [Quit: Bridge terminating on SIGTERM]
hypokeimenon[m] has quit [Quit: Bridge terminating on SIGTERM]
shu9 has quit [Quit: Bridge terminating on SIGTERM]
josias has quit [Quit: Bridge terminating on SIGTERM]
colemickens has quit [Quit: Bridge terminating on SIGTERM]
M011100110110110 has quit [Quit: Bridge terminating on SIGTERM]
danielrf[m] has quit [Quit: Bridge terminating on SIGTERM]
katrms[m] has quit [Quit: Bridge terminating on SIGTERM]
Yakulu[m] has quit [Quit: Bridge terminating on SIGTERM]
ncm[m] has quit [Quit: Bridge terminating on SIGTERM]
hiroshi[m] has quit [Quit: Bridge terminating on SIGTERM]
siraben has quit [Quit: Bridge terminating on SIGTERM]
rhysmd has quit [Quit: Bridge terminating on SIGTERM]
packetup has quit [Quit: Bridge terminating on SIGTERM]
kenmacd[m] has quit [Quit: Bridge terminating on SIGTERM]
thefloweringash has quit [Quit: Bridge terminating on SIGTERM]
superherointj[m] has quit [Quit: Bridge terminating on SIGTERM]
Lestat[m] has quit [Quit: Bridge terminating on SIGTERM]
Ox4A6F has quit [Quit: Bridge terminating on SIGTERM]
cryptix has quit [Quit: Bridge terminating on SIGTERM]
philipp[m]1 has quit [Quit: Bridge terminating on SIGTERM]
colemickens has joined #spectrum
josias has joined #spectrum
siraben has joined #spectrum
danielrf[m] has joined #spectrum
philipp[m]1 has joined #spectrum
shu9 has joined #spectrum
mvnetbiz_ has joined #spectrum
M011100110110110 has joined #spectrum
Lestat[m] has joined #spectrum
Yakulu[m] has joined #spectrum
ncm[m] has joined #spectrum
hypokeimenon[m] has joined #spectrum
packetup has joined #spectrum
Ox4A6F has joined #spectrum
thefloweringash has joined #spectrum
hiroshi[m] has joined #spectrum
katrms[m] has joined #spectrum
superherointj[m] has joined #spectrum
enick_4 has joined #spectrum
kenmacd[m] has joined #spectrum
cryptix has joined #spectrum
rhysmd has joined #spectrum
<Profpatsch>
Did somebody send a kill to -1 on the matrix bridge server
<qyliss>
With virtio wayland you can (optionally) use the host gpu, fwiw
manveru has quit []
manveru has joined #spectrum
<v0idify>
is it possible to isolate hardware into VMs like VT-d on aarch64 or other platforms?
<TheJollyRoger>
v0idify: POWER has POWER IOMMU, which is even better, but if I recall correctly, enabling it in strict mode will cause some things to break.
<TheJollyRoger>
Since our programmers and developers have been making x86-specific assumptions for decades now.
<TheJollyRoger>
ARM/Aarch64 has SMMU which is analogous to VT-d.
<TheJollyRoger>
But it's up to the vendor to implement and use it.