🏴☠️ 𐕣 V A L ░ E R I E 𐕣 🏳️⚧️trying to address this resurfaced rubberbanding on my opensim grid, apparently maybe has to do with bulletsim, i'm not quite sure
this all makes me wonder what kind of architecture mainline Second Life uses for its regions, because it definitely can't be much bigger than what I'm using for my 8 simulators, although I wonder if they're using 1 machine per simulator
also 1 machine per simulator would be crazy lol they'd have to have a server farm miles long
although, that probably is how they do it, because every time you switch regions in SL, it acts as though you're moving to a different server by reporting such in the chat window
huh
what if i offload the heaviest regions to another server
Meadow Whisper (Natasha L.)@valerie That's exactly how SL does it, and it's partly why they moved everything to AWS (which backfired spectacularly because Amazon doesn't have anywhere near the same connection reliability between servers that LL had in their own datacenter, and that's why region crossings got a lot worse after they made that switch). They share things a bit: Full-size regions are 4 per server, and Homesteads are 16 (I think, it's been a while), but that's on *huge* hardware allocations. For anything any normal person would have at home or rent from a VPS host, assume one server per region
No wonder I’m experiencing rubber banding with 28 regions across 8 simulators on 1 server haha
But I’m going to try to allocate 1 core per simulator, that might fix it. OpenSim/SL is pretty old software, no way it needs more than 1 core per simulator
@valerie It may be old, but the big issue is latency: Modern virtualization takes a lot of shortcuts to split hardware between systems, and a sort of "just in time" approach, but the simulator code for SL and OpenSim relies on immediate access to sending CPU instructions and getting results. When it has to wait even a fraction of a microsecond, it triggers a cascade of incorrect physics calculations that can quickly build up to become visible to connected users. This was a massive problem when they moved to AWS too; the CPU requirements per simulator aren't that much, but the CPU *latency* requirements make it extremely difficult to virtualize reliably.
At a minimum, you'll need a dedicated CPU core per simulator that isn't also doing stuff for the OS or virtualization environment.
Yeah, I figured that’s probably what I need. Right now all cores are sharing everything so when one simulator hiccups, everything does
stefan (シュテファン)@valerie @lupinia Quick question: How many simulators are you running in #opensim and how many regions does each one have?
I’m asking so that, as a newbie, I can get a rough idea of what you’re talking about. I’m currently planning a simulation with 9 regions in two different sizes, 1024 and 2048. Now I’m wondering whether this issue could affect me as well.
I’m running 8 simulators with 4 512 regions per simulator, and once I give each sim their own thread it should perform as well as second life, I wouldn’t load up a simulator with more than that though I think