We tried to simulate 1000 viewers on a #PeerTube video, and then on a livestream. (Note that 99% of twitch livestreams are under 1000 simultaneous views.)
We optimized accordingly... And here are the results: it works!
With a ~20$ server, PeerTube can now support such usecases.
Read all about these stress tests on our blog : https://joinpeertube.org/news/stress-test-2023
@peertube ... I am this close to actually try to setup a peertube server now .
@peertube super interesting! What would you say is the bottleneck for a VOD video gone viral across multiple federated instances in addition to the origin? Would the concurrent viewer count be still limited by the origin in some way, or is the video watchable as long as viewers are balanced equally so as to not overload a single instance?
@peertube
Il y a une typo : "800MB téléchargés/100MB téléchargés".
Les 800 sont en envoi.
@John_Livingston Corrigé, merci :) @peertube
@spf @John_Livingston @peertube Pour la partie Vidéo VOD le texte dis "C'est pourquoi nous avons un pic de 260Mbit/s pour le téléchargement HTTP au début du graphique (15:09:00)." J'ai l'impression qu'on est plus vers les 206Mbit/s, c'est aussi une typo ou je regarde pas au bon endroit ? Bon après ça fait pas une grosse différence . Merci pour ces tests et cet article
@haughtgard Alors pour le coup, j'ai fait passer : je n'ai pas écrit la news (j'ai juste corrigé la coquille) :)
@haughtgard @spf @John_Livingston @peertube Merci je corrige !
@peertube PeerTube is single threaded (well, node is), is there a point in giving it more than 2 vCPUs?
I guess it's for transcoding, but with remote runners you could start a big ass VM on demand. It would help reduce costs.
@Arcaik @peertube PeerTube uses multiple cores. NodeJS can use multiple threads using worker threads: https://nodejs.org/api/worker_threads.html
@Chocobozzz @peertube Oh ok.
From the results above it seemed consistent with my experience working with single threaded node applications. Basically CPU never goes above 120-150% (1 core for the event loop and a little bit more for node's “internal” threads).
@peertube Neat! fyi, it looks like the blog posts have the "og:url" metadata pointing to the joinpeertube[.]org root page, so that's what the rendered card ends up linking to, rather than the blog post itself
@peertube nothing beats real world tests but I wonder how close they were to local dev simulations or estimates.
Those are kickass numbers! I do wonder if PT could handle the stress of the top few streamers of enough hardware resources were available... I bet it felt ultra powerful generating 1000 IPs heh like raising an zombie army.
I can't wait to update to 6.2+ but I do hope the update process smoothes out more over time... My docker based PaaS is def not making the process any easier lol.
@Tornuggla @peertube @janboehm @stefan Der Preisfaktor bei PeerTube ist der Speicherplatz auf der Festplatte. Ein ZMR-Kanal würde 200 GB verbraten, und wenn mehrere Auflösungen angeboten werden sollen, schnell das doppelte.
haha, das Live-Labern von @janboehm ist ja jedenfalls nur in Handy-Auflösung, aber wenn er das wöchentlich macht kommt da schon was zusammen...
@Tornuggla Die machen auf YouTube jetzt neuerdings 4k, da kommt noch einiges an Datenmengen zusammen.
Ein weiterer Punkt, der zu beachten wäre bei einem Umzug nach PeerTube ist die fucking Depublikationspflicht beim #ÖRR. Da muss zuerst einmal die Rechtsabteilung vom @ZDF ran um zu prüfen, ob das überhaupt geht.
@peertube Does this work with multiple quality levels? Show-stopping flaw in Twitch is that you can't watch without extremely high bw. Sorry anyone on DSL, metered cellular, etc. It'd be awesome if PeerTube has solved this.