#pest | 2023-09-21

I would like to know what kind of protocol can be built on top of pest, and which can't.

↖

16:26

not clear on why you're choosing to work with an encoding with worse time complexity but I'm not gonna stop ya

16:30

Not really worrying about that yet, I can defer that decision for now, signpost.

16:32

I see that the Maymounkov-Mazieres' protocol require to interact each time the peer table is updated. I dunno if pest was tested in roaming situation (for example train).

↖

16:34

The other encoding is more like p2p where you request an n-bytes file and the sender computes n+r-bytes stream, with r the estimated packet loss (I dunno if it can be guessed based on pest traffic).

16:38

From my understanding, the only thing a pest station may guess from recent messages is that it can reach a certain peer through a list of peers, and spread the messages to them to increase chances.

16:48

l <= m, m is not the message size, but the exponent in F_{q^m}. Will try to simplify a bit and make an example.

16:53

There are so many layers of vector spaces I can't be more precise right now heh

16:59

signpost: do you have an idea of what a transfer protocol could look like? Because from what it looks like, bdl looks like there are seeders that broadcast available files, and once selected it start sending stuff until it reaches the end, and eventually restart sending missing stuff.

17:03

*

unpx reminds himself to read logs http://logs.nosuchlabs.com/log/pest/2022-07-25#1010640

17:03

dulapbot

Logged on 2022-07-25 23:35:28 crtdaydreams[asciilifeform]: http://logs.nosuchlabs.com/log/pest/2022-07-25#1010637 << when mentioned earlier, flamed

~ 20 minutes ~

17:23

To me, the bdl requires a dht, so I'm not sure how this could be adapted to pest. But I will ponder.

~ 32 minutes ~

17:56

http://logs.bitdash.io/pest/2023-09-21#1030364 << this was an explicit design goal (see sect. 1.7 of spec) and in fact tested, iirc phf in fact at least once was speaking from actual train, via gsm box

↖

17:56

bitbot

Logged on 2023-09-21 16:32:33 unpx[jonsykkel|signpost]: I see that the Maymounkov-Mazieres' protocol require to interact each time the peer table is updated. I dunno if pest was tested in roaming situation (for example train).

17:58

re p2p warez, the notion is that you ask a set of peers to start fountaining at you, and they do so for a certain time, after which -- if you haven't the pieces you need -- you ask'em again, and so forth until you have'em

17:59

the moar familiar scenario where you have nfi who exactly you're asking, and who you expect will answer, is not applicable in pest

18:02

something like who-haz $hash; i-haz $hash; pls-send-up-to $hash $blockCount

18:02

and you can request blocks from whichever of those responded in the affirmative

18:02

and of course they can decide they dun wanna, and you just request more blocks from the willing.

18:03

iirc maymounkov et al came up with a more complicated scheme; I've forgotten now, but I don't see a need for something more complicated than the above.

~ 2 hours 10 minutes ~

20:14

http://logs.bitdash.io/pest/2023-09-21#1030361 << in general, the only sort of protocol that you defo ~can't~ build on pest, is the kind which requires talking to strangers

20:14

bitbot

Logged on 2023-09-21 16:25:55 unpx[jonsykkel|signpost]: I would like to know what kind of protocol can be built on top of pest, and which can't.

20:14