#asciilifeform | 2021-10-04

02:32

punkman: https://pdos.csail.mit.edu/~petar/papers/maymounkov-bigdown.pdf << highly recommended reading

02:33

once you're talking about blasts of more than a single item, questions like "wat do when I lose something" come in, and imho the linked item points to a much better approach than reimplementing tcp shittily

↖

02:33

*

signpost has a WIP item that may bolt nicely to thimbronion's before long, in py

↖

02:35

signpost

one lovely consequence of the encoding stream is that, for items multiple peers have, multiple peers can start spraying blocks at you, and you'll reap the benefit of the parallelism without them coordinating with each other on what to send you.

02:35

*

signpost recs as many folks as are interested in this to read the paper, and then the more math-heavy one.

02:35

*

asciilifeform not fond of multipacket answers to a 1-packet question, for the simple reason where 'am i expecting this packet?' gotta be O(1) resolvable

↖

02:35

dulapbot

Logged on 2021-10-01 14:12:28 asciilifeform: punkman: sadly asciilifeform not had time to write this section ! and doesn't know when will. intention was 1 hash per msg tho, to make easier calculation of the cpu cost of processing a valid packet.

02:37

02:32	signpost	punkman: https://pdos.csail.mit.edu/~petar/papers/maymounkov-bigdown.pdf << highly recommended reading
		↖
02:33	signpost	once you're talking about blasts of more than a single item, questions like "wat do when I lose something" come in, and imho the linked item points to a much better approach than reimplementing tcp shittily
		↖
02:33	*	signpost has a WIP item that may bolt nicely to thimbronion's before long, in py
		↖
02:35	signpost	one lovely consequence of the encoding stream is that, for items multiple peers have, multiple peers can start spraying blocks at you, and you'll reap the benefit of the parallelism without them coordinating with each other on what to send you.
02:35	*	signpost recs as many folks as are interested in this to read the paper, and then the more math-heavy one.
02:35	*	asciilifeform not fond of multipacket answers to a 1-packet question, for the simple reason where 'am i expecting this packet?' gotta be O(1) resolvable
		↖
02:35	dulapbot	Logged on 2021-10-01 14:12:28 asciilifeform: punkman: sadly asciilifeform not had time to write this section ! and doesn't know when will. intention was 1 hash per msg tho, to make easier calculation of the cpu cost of processing a valid packet.
02:37	asciilifeform	signpost: imho lubyism is a++ for pest-style known-peers net, as the need to separately authenticate the frags falls away (i.e. can generally trust that a direct peer is not sending garbage frag)
02:37	asciilifeform	for warez application -- perfect
02:38	signpost	exactly. I'm saying punkman does not need this for base-layer pest, and pointing to where it has an application.
02:38	asciilifeform	(for chat, not really needed, but simple to bolt on as a command code)
02:38	asciilifeform	aha
		~ 2 hours 54 minutes ~
05:32	punkman	http://logs.nosuchlabs.com/log/asciilifeform/2021-10-03#1060616 << if some packets are lost, we would simply send more getdata requests.
05:32	dulapbot	Logged on 2021-10-03 22:33:25 signpost: once you're talking about blasts of more than a single item, questions like "wat do when I lose something" come in, and imho the linked item points to a much better approach than reimplementing tcp shittily
05:35	punkman	http://logs.nosuchlabs.com/log/asciilifeform/2021-10-03#1060620 << if peer gets request to fill 5k message gap, could decide to only send 10 or 50, and requester will eventually send more getdata requests
		↖
05:35	dulapbot	Logged on 2021-10-03 22:35:27 asciilifeform: not fond of multipacket answers to a 1-packet question, for the simple reason where 'am i expecting this packet?' gotta be O(1) resolvable
05:36	punkman	it does add unwanted complexity, will concede the point
		~ 2 hours 3 minutes ~
07:39	punkman	in other complexities, what about compressing messages? would help avoid splitting of ~500 byte irc messages into 2 pest messages. Relevant: https://github.com/facebook/zstd#the-case-for-small-data-compression
		↖
		~ 6 hours 19 minutes ~
13:59	signpost	pest seems like a thing to implement as simply as possible, then see what problems one actually has
		~ 38 minutes ~
14:37	asciilifeform	^
14:39	asciilifeform	http://logs.nosuchlabs.com/log/asciilifeform/2021-10-04#1060632 << imho this is an absolutely terrible idea. introduced 'over9000' gnarly moving parts, for dubious gain. (i dunthink there has ever existed a compressor w/out vulns. at the very least, the basic 'zip bomb' afflicts them all. and go and try to calculate, w/out burning cpu, how much it costs to unzip a compressed turd..)
14:39	dulapbot	Logged on 2021-10-04 03:39:49 punkman: in other complexities, what about compressing messages? would help avoid splitting of ~500 byte irc messages into 2 pest messages. Relevant: https://github.com/facebook/zstd#the-case-for-small-data-compression
14:39	asciilifeform	*introduces
14:40	asciilifeform	http://logs.nosuchlabs.com/log/asciilifeform/2021-10-04#1060629 << i'd really NOT like to re-create what imho is the single most dysfunctional aspect of trb, in pest...
14:40	dulapbot	Logged on 2021-10-04 01:35:59 punkman: http://logs.nosuchlabs.com/log/asciilifeform/2021-10-03#1060620 << if peer gets request to fill 5k message gap, could decide to only send 10 or 50, and requester will eventually send more getdata requests
14:41	asciilifeform	$ticker btc usd
14:41	busybot	Current BTC price in USD: $47682.26
14:41	asciilifeform	!w poll
14:41	watchglass	Polling 17 nodes...
14:41	watchglass	185.85.38.54:8333 : Could not connect!