Show Idle (>14 d.) Chans
← 2022-09-19 | 2022-09-21 → ↓
11:52 awt $ticker btc usd
11:54 asciilifeform !q seen busybot
11:54 dulapbot busybot last seen here on 2021-11-18 09:41:04: PeterL: who is your daddy and what does he do?
11:55 asciilifeform hrm nm mine hasn't the foo[bar] remover
11:55 asciilifeform !q seen busybot[asciilifeform]
11:55 dulapbot busybot[asciilifeform] last seen here on 2022-09-13 11:52:37: Current BTC price in USD: $20835.34
~ 2 hours 21 minutes ~
14:16 * asciilifeform apropos of nuffin, as of today 30yrs of sitting in usa, lol
~ 1 hours 2 minutes ~
15:18 awt happy intake day
~ 40 minutes ~
15:59 asciilifeform lolty
16:11 PeterL what are we taking in?
~ 25 minutes ~
16:37 awt PeterL: asciilifeform went through USG intake process 30 years ago today, it seems.
16:37 awt https://www.quora.com/What-is-intake-like-in-prison
~ 1 hours 53 minutes ~
18:30 jonsykkel asciilifeform: looks good, as far as i can tell u would expect that if both X+Y are live in the net, they will "always" be able to recv addrcasts from each other, and at aprox the same time. but if T_p is comfigured diferently they will start hammering out of sync (also wat happens after H has expired? give up or retrry la
↖
18:30 jonsykkel ter)
18:30 jonsykkel thing 2: not 100%sure my internal model of nat is corectt but wouldnt the maffs depend hevily on exact nature of nat? at least my understanding of wat the problem looks something like this:
18:31 jonsykkel unpredictable (somewat) factors: nat maping table size (no idea, according to some random account, on order of 2^10-2^14 for typical consoomer router, unlikely to be significantly smaller than that 2^10 at least), timeout of the mapings (usually in the range of 30sec to couple minutes max from wat i can figure)
18:31 jonsykkel so if we assume "worst case", 30sec timeout, either the table size or the spam interval will be the limiting factor. u prolly dont want to be spamming faster than something like 0.1ms
18:31 jonsykkel which means u will have a "moving window" (as rules time out and get replaced with new ones) of at most 30sec/0.1ms = 300k entries (would guess the tables tend to be smaller than this in practice). lets call window size "W" and spam interval "S"
18:31 jonsykkel in the case of the maximally annoying nats, each entry in table wil have src+dst ports that both have to match
18:31 jonsykkel which means, after this window has been "saturated" (W*S sec has passed), u have W/(64512^2) = between 0.000025% (W=2^10) and 0.0072% (W=300k) chance of succes for every packet that actually arrives
18:31 jonsykkel this translates to chance of success (assuming 0%packet loss) within 60sec = 1-((1-W/(64512^2))^(2*(60/S))) = 25.6% (W=2^10) or ~100% (W=300k)
18:31 jonsykkel both my nats are definitely of the annoying type, confirmed by experiments (can not send to nat assigned port from the same external host throguh a diffrent udpsocket). the litle data i colected seems to indicate my W might be somewhere in 3k-30k range
↖
~ 18 minutes ~
18:49 jonsykkel also, if retry later after H expired - how to sync?
18:50 jonsykkel supose could have flag in the addres casts "i will start hammering T_p sec after sending this packet, if prods fail. plox to do same"
↖
18:51 jonsykkel then flag would be set for initial addrcast and then every x min or watever
19:03 jonsykkel actualy not sure why i thouhgt "always at arpox same time", thers no guarante of this at all. but i gues the H would be much longer than the adr cast interval
19:03 bitbot Logged on 2022-09-20 18:30:50 jonsykkel: asciilifeform: looks good, as far as i can tell u would expect that if both X+Y are live in the net, they will "always" be able to recv addrcasts from each other, and at aprox the same time. but if T_p is comfigured diferently they will start hammering out of sync (also wat happens after H has expired? give up or retrry la
~ 46 minutes ~
19:50 asciilifeform jonsykkel: if operators mostly agree on a default T_c, then hammer will start at roughly same time.
↖
19:51 asciilifeform ( if not, then connection will take multiple hammerings, initiated by >1 addrcast )
19:51 asciilifeform eventually they'll connect.
19:52 asciilifeform ditto T_p etc.
19:53 asciilifeform afaik there are no NATs where it'll start (i.e. lets out udp at all) where hammer won't eventually penetrate.
19:55 asciilifeform old-style (pre-microshit) skype used this algo.
19:58 asciilifeform http://logs.bitdash.io/pest/2022-09-20#1013285 << any idea whether the ephemeral port #s predictable in these ? imho worth a look
↖
19:58 bitbot Logged on 2022-09-20 18:31:42 jonsykkel: both my nats are definitely of the annoying type, confirmed by experiments (can not send to nat assigned port from the same external host throguh a diffrent udpsocket). the litle data i colected seems to indicate my W might be somewhere in 3k-30k range
19:58 bitbot Logged on 2022-09-19 18:34:53 asciilifeform[4]: suspects that the hammer time can be shortened considerably by preferentially exploring the space around the ephemeral ports seen in successfully-received prods from currently-warm peers -- many NATs issue ephemeral ports sequentially
20:00 asciilifeform http://logs.bitdash.io/pest/2022-09-20#1013287 << not bad idea imho, but possibly not needed if general agreement re intervals
20:00 bitbot Logged on 2022-09-20 18:50:57 jonsykkel: supose could have flag in the addres casts "i will start hammering T_p sec after sending this packet, if prods fail. plox to do same"
~ 47 minutes ~
20:48 jonsykkel http://logs.bitdash.io/pest/2022-09-20#1013291 << but common case will prolly be that one guy restarts station rather than working peering going cold while both stations running
20:48 bitbot Logged on 2022-09-20 19:50:37 asciilifeform[5]: jonsykkel: if operators mostly agree on a default T_c, then hammer will start at roughly same time.
20:50 jonsykkel so, if X restarts station, X and Y are out of sync re coldness of eachother
↖
20:52 jonsykkel perhaps can be fixed by simply responding immediately to a received addr cast IF the last one u sent to this peer was long ago
↖
20:55 jonsykkel ttp://logs.bitdash.io/pest/2022-09-20#1013297 << one nat was 4g carriers, the other one some router
20:55 jonsykkel http://logs.bitdash.io/pest/2022-09-20#1013297 << one nat was 4g carriers, the other one some router
20:55 bitbot Logged on 2022-09-20 19:58:38 asciilifeform[6]: http://logs.bitdash.io/pest/2022-09-20#1013285 << any idea whether the ephemeral port #s predictable in these ? imho worth a look
20:55 jonsykkel 4g one appeared entirely random every time from the full range
20:56 jonsykkel the router seemed to assign ports clustered around the same place, but i dont know whether that was cuz src port was the same, or cuz mappings created around same time
20:56 jonsykkel ill try to figure out the logic, but presumably they all work diffrently
~ 19 minutes ~
21:16 jonsykkel http://logs.bitdash.io/pest/2022-09-20#1013304 << or rather, their casts will be out of phase. worst case one guy starts hammering T_a before the other
21:16 bitbot Logged on 2022-09-20 20:50:32 jonsykkel: so, if X restarts station, X and Y are out of sync re coldness of eachother
~ 2 hours 10 minutes ~
23:27 crtdaydreams pestron borked, aborting
23:27 jonsykkel ugota update to 96K so it knows wat to do with adres cast
23:28 jonsykkel or prod rather
~ 18 minutes ~
23:46 crtdaydreams generally, what do most folks have their address cast interval set to?
23:47 crtdaydreams also, unsure if bug or feature, but getting ~minimum~ 5 AC packets from all peers in short interval
23:48 crtdaydreams jonsykkel: tried increasing acint but still making 5 AC requests by the looks of it, which chewing up bandwidth, pestron making the requests almost every 10 seconds
23:50 jonsykkel crtdaydreams: knob works, u might be seing rebroadcasts of incoming ACs
23:51 jonsykkel theres tons of these
↖
23:53 crtdaydreams jonsykkel: hm it seems a bit excessive
23:54 crtdaydreams whats log output on your end look like? is my pestron blasting AC over9000 times a minute, also 10x'ed my prodint.
23:55 jonsykkel crtdaydreams: http://zzz.st/up/E7Riuzjx/
23:56 jonsykkel u see outgoing AC hash matches incoming one, aka rebroadcast
23:57 jonsykkel exept those last4, which are from my statoin
23:57 * crtdaydreams recalls jonsykkels pentagram diagram of net broadcasts
23:58 crtdaydreams herm
23:59 crtdaydreams exponentially proportional to size of net
23:59 jonsykkel u also gota set igint 10000 or smth if wish to minimize bw
23:59 crtdaydreams what igint do?
23:59 jonsykkel ignore pakets inteval
23:59 crtdaydreams ah
← 2022-09-19 | 2022-09-21 → ↑

Random(pest) | Download hourly DB snapshot | Get Source Code

Theme: Light | Dark | System Default