Jendrik | JeWe37

CES @RWTH. Here: Mostly Politics and Philosophy. Former tech MC player.

[Jendrik | JeWe37 @JeWe37]

@VictorTaelin @jmbollenbacher net negative, become net positive.

we could test population wide, but all that that would do is leave us conducting expensive tests before checking for other symptoms that would actually make those tests' outcomes actionable.

[Jendrik | JeWe37 @JeWe37]

@VictorTaelin @jmbollenbacher still incorrect! a test, which is conducted after a patient showed prior symptoms, will leave us in a situation wherein we have higher confidence they actually have the disease we tested for than when testing an asmyptomatic patient. consequently, outcomes that would have been

[Jendrik | JeWe37 @JeWe37]

@yashkaf what. if a straw has 1 hole a tshirt clearly has 3 what are these options

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev it s just XML right? so can't you just use sth like xq(jq but with XML to json in front)?

[Jendrik | JeWe37 @JeWe37]

@inerati this account used to be 100% of the finest bait. the world really has gone so mad that contrarian takes are now but the sanest things in the world huh.

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev more sense is to think about shutting down the container when not in use. using postgres kinda mostly has the advantage that if you re only aggregating state its already there, but if you re using it for something directly, you might as well have it be aggregated where it's used.

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev necessarily be impossible either, especially when you consider "jobs" really aren't really the right way to look at this when you consider a single GPU can usually easily transcode several streams yet you would typically be allocated a single GPU per container. so what would make

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev by just sending each user directly to a different transcoder. this also means the playlist manager has all the information it would need to report for playback state and similar features, and since it doesn't do much, centralizing it doesn't hurt.

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev gives it all the information required to serve multi variant playlists that make sure that the individual variant playlist urls point to the transcoder which either A) already has segments for the file or B) is under the least load. that way you can manually do load balancing

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev playlists and segments. that makes aggregation of information relatively simple, while also giving you full control over the load balancing without being overly dependent on complex configuration of a specific load balancer, like would be needed for watch together.

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev I see, but one could also imagine centralizing the computationally less demanding but relatively interdependent workload of serving the multi variant master playlist, while then creating workers only for serving and doing the required transcoding of the individual variant

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev sure, but thats far simpler to implement than attempting to work with centralized storage for distributed workers, isnt it? probably boils down to a setting in the loadbalancer and perhaps making sure that the playlists served by each worker actually only point to that worker

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev stream would require more than a single transcoder to get up to 1.0x speed, but at that point stream start/scrubbing latency would be borderline unworkable. and this would be a substantial effort to architect an implement by comparison

[Jendrik | JeWe37 @JeWe37]

@zoriya_dev how would this be advantageous compared to just sending the user to an entirely separate transcoder instance(that doesn't happen to have segments for the requested stream on hand) depending on which is currently under the least load? this seems only useful when a single transcode

[Jendrik | JeWe37 @JeWe37]

@xsphi @keysmashbandit now just sum from n=0 to 99 to get the number of expected loops

where the heck do i go wrong here cause that seems far too easy(and trivial to extend to the other case) and gives me 3.284 loops which also agrees with my simulations?

[Jendrik | JeWe37 @JeWe37]

@xsphi @keysmashbandit you are taking exactly 100 steps(200 ends, each step removes two thereof). after picking your first end, where which you pick has no effect, the second end is the one that would form a loop with a probability of 1/(199-2n) because after n steps there are only 199-2n ends left.