This message was deleted.

@steep-student-24587 pls thread your messages. Can you repost your screenshots here?

1: During load testing, I noticed that one meeting stops at 75/75 participants and then another page starts. Could you help me understand why this transition occurs? 2: During the load test, I observed that 9999 subscribers joined the room. Is this expected behavior, or does this number seem unusually high? Are these subscribers actively interacting with the meeting content, or do their interactions seem abnormal in any way? 3: After stopping a meeting during load testing, I obtained certain results. here is the result @magnificent-art-43333

what is your workload? how many publishers and how many subscribers? your load tester is seeing high packet loss, which means you've exceeded either the CPU or network capacity of either the SFU or the load tester client.

my workload is for 1 publishers and 10000 subscribers.

Have you taken a look at the OSS benchmarks here? https://docs.livekit.io/oss/benchmark/

yeah. i did the load test with help of this only. i cam see 3k sub for 1pub taking 92% CPU. but i tetsted it with autoscaling for 1 room.

But you’re trying 10K subs on a similar setup?

yes but with scaling.. when CPU utilization is on 50% it will spin up another node.

LiveKit Open Source doesn’t mesh/cascade, so a single room can only be hosted on a single node/machine.

yes, but can the users/subs join the room from another nodes/machine?

No. With LK Open Source, everyone joining the same room connects to the same machine.

okey. so no scaling will work in case of LK opensource?

If you have a 16-core server and your subs grow beyond 3k, I’d say that’s a great problem to have. 😊 You can vertically scale a room with LK open source, but horizontal scaling is not included out of the box. You can certainly build it yourself, and there’s a few different ways to do it depending on your needs. But, unless you’re already regularly doing 3K or 10K subscriber rooms (again, a fantastic problem to have), my recommendation is you start with LK Cloud. We’ve already solved the horizontal scaling problem there. If you end up consistently running really big rooms, you can ping @bright-scooter-10854 to chat through options. If you ever decide you want to self-host down the road, the APIs/SDKs are identical.

how can we achive vertically scale a room with LK open source, cAN you please help me with this

I’m far from an expert here, but I don’t think you can really do this on the fly. Your scale is dependent on the server you select to run LK OSS. A 32-core machine and 10gbps link will be able to handle a larger room than a 16-core and 2gbps. As I mentioned above, unless you already have a product at this scale (10K user streams), I recommend you start with LK Cloud so you can build your product more quickly, start testing it with users and figure out if there’s initial product-market fit, then figure out how to deploy LK OSS if you want to go that route. If you’re already running this scale or your product/project grows so quickly your servers start to melt, ping @bright-scooter-10854 and we can figure out how to help.