asciilifeform: this is a temporary peace.

asciilifeform: BingoBoingo: they're still pestilential

asciilifeform: fragola makes this impossible in principle, you gotta stow N frags to get any sense of whether the full packet is friend or foe

asciilifeform: the only final solution to ddos is O(1) crapolade packet rejection. ( preferably in iron )

asciilifeform: of course ddosers can send packets with your new protocol # , it doesn't make you bulletproof

asciilifeform: Mocky: incidentally, it is possible to do what i suggested to him then, which is to change the protocol # in the ip header and be generic ip, rather than udp . but only once we have own ip stack.

asciilifeform: Mocky: i suspect that he's rtfm'd since

asciilifeform: Mocky: my suspicion is that under ddos conditions, loss of frags will approach 100%

asciilifeform: i'll observe that if you dispense with fraggism, ip stack becomes fairly easy to write ( i.e. if you support only unfragged udp and nuffin else )

asciilifeform: the ip stack's frag/reasm is one of those things that 'worx until it doesnt'

asciilifeform: imho if you want large messages, oughta have own fragger/reasmer, not the ??? in linux/ciscolade ☟︎

asciilifeform: Mocky: me neither, esp. given as it wins nuffin bandwidth-wise

asciilifeform bbl,teatime

asciilifeform: i for one would rather have no frag reassembly at all if writing ip stack. not only b/c complexity but also this. ☟︎

asciilifeform: which not only complicates ip stack ( for when we write one ) but opens up to ddosability ( frags are take-it-or-leave-it, they dun even carry the port # )

asciilifeform: i.e. it is fundamentally an 'allcomers-can-claim-some-ram' item

asciilifeform: even if seems that 100% of 2/3-frag packets make it through in 'laboratory' conditions, still gotta remember that the frag reassembly buffer is the ~exact~ equivalent of the pre-trb 'block orphanage' ☟︎

asciilifeform: re : udpism : at the risk of rehashing some of the ancient gossipd thread, i'ma put a few notes re fragging :

asciilifeform: mod6: mostly mired in saecular liquishit

asciilifeform: slowly.

asciilifeform: ohai mod6

asciilifeform: diana_coman: http://btcbase.org/log/2018-09-28#1855195 << output ☝︎

asciilifeform: diana_coman: 10ms

asciilifeform: diana_coman: all of my testfires thus far ended up 'no loss, no reorder, as if on lan'

asciilifeform: http://btcbase.org/log/2018-09-28#1855291 << lemme know if you'd like sumthing in particular tested with my path, diana_coman ☝︎

asciilifeform: !#s octeon

asciilifeform: http://btcbase.org/log/2018-09-28#1855180 << also forgot to answer -- it's a mips 'octeon' running freebsd off ssd , iirc described in several old thrds ☝︎

asciilifeform: http://btcbase.org/log/2018-09-28#1855188 << nitpick: >1500byte always fragged, cuz ethernet. but! apparently get sewn back together in time. at least at the currently tried rates, and with mix of sizes ( remains to be seen what receiver will do with a summed MB/s of frags from different people ) ☝︎

asciilifeform: mircea_popescu: it not only made for very picturesque output in old buggy vtrons, but pretty terrible for blood pressure, as turned out that the supposed 'disable fuzzy' flags dun actually do anyffin in gnupatch

asciilifeform: mircea_popescu: shouldn't take much sweat, anyffing that calls gnupatch could just as readily call phf's

asciilifeform: indeed, 0 nitrates!1111

asciilifeform: ( recall, we had gpg-signed patches with 0 robotics for yr+ )

asciilifeform: mircea_popescu: lol recall how we even ended up with v, ' asciilifeform : 'it is obvious!11 how to arrange trb patches' errybodyelse : 'nah' )

asciilifeform sings...

asciilifeform: ahahahayes

asciilifeform: lol

asciilifeform: mircea_popescu: phf posted one earlier

asciilifeform: they dunwork. cuz how would they.

asciilifeform: any an' all of the 'soft' hacks

asciilifeform: esp on os with liquishit scheduler

asciilifeform: ~iron~ mixers : work. 'soft' liquishit -- surprise, surprise -- doesn't

asciilifeform: THEN stopped when 'winmodem' soundcards.

asciilifeform: mircea_popescu: phunphakt : mixer worked GREAT 20y ago. when hardware dsp in sound blaster.

asciilifeform: ( spoiler : didn't )

asciilifeform: iirc this was even poettering's orig worming-in -- he claimed to fix mixing

asciilifeform: last i knew.

asciilifeform: i dun know the specific answer. but suspect it has to do with the sad audio mixer on most os. they wanna hear the game sounds + the chat.

asciilifeform: i.e. with telephones ?

asciilifeform never saw the appeal

asciilifeform: i meant those mud folx who have microphones in the game

asciilifeform: i've nfi why you'd want >1k/s per user, unless you were doing voices or somesuch exotica

asciilifeform: in battlefield -- definitely

asciilifeform: will be interesting to try a shot with several people txing from different places. see if it triggers antiddos derpery somewhere.

asciilifeform: ( even if erry client sends at 300 baud )

asciilifeform: mircea_popescu: keep in mind that traffic on receiver will be considerably moar than 1k/s

asciilifeform: mircea_popescu: then yer golden, loox like. at least if errybody has a path no worse than mine

asciilifeform: mod6: i seem to recall a much sadder london test but that was with very heavy packets iirc

asciilifeform: video does get to skip frames & buffer etc

asciilifeform: 10ms is still pretty relaxed pace tho.

asciilifeform: 0 reorderings too, loox like

asciilifeform: apparently

asciilifeform: mircea_popescu: i suspect wasn't even in '94

asciilifeform: ( still in 1 direction )

asciilifeform: 3) http://nosuchlabs.com/pub/udpism/usa_tx_10ms_run3.txt http://nosuchlabs.com/pub/udpism/uy_rcv_10ms_run3.txt

asciilifeform: 2) http://nosuchlabs.com/pub/udpism/usa_tx_10ms_run2.txt http://nosuchlabs.com/pub/udpism/uy_rcv_10ms_run2.txt

asciilifeform: 1) http://nosuchlabs.com/pub/udpism/usa_tx_10ms_run1.txt http://nosuchlabs.com/pub/udpism/uy_rcv_10ms_run1.txt

asciilifeform: here goes : ☟︎

asciilifeform: yea i suspect 508 is a textbookism

asciilifeform: about to post the 10ms variant ( 3 shots )..

asciilifeform: thus far they all show up on my exit router, fwiw

asciilifeform: ( i dun have a tap in florida, lol )

asciilifeform: how does one determine exactly which hop

asciilifeform: i'ma try with 10ms

asciilifeform: ( afaik 1sec is way moar than long enuff for a packet to either make it, or vanish )

asciilifeform: upstack, re the udp experiment -- 1/sec is sorta 'cheating', no possibility of reorders

asciilifeform: worx a++

asciilifeform: mircea_popescu: no disagreement re upgrades of parts

asciilifeform: i.e. ->, <-, ->, ...

asciilifeform: ( potentially also could be interesting to make echo variant )

asciilifeform: mircea_popescu: so far only tried the -->

asciilifeform: i.e. : cat uy_receiver_udp_log.txt | cut -f 1 -d ',' > receiver_idx.txt ; cat usa_sender_udp_log.txt | cut -f 1 -d ',' > sender_idx.txt ; diff sender_idx.txt receiver_idx.txt << produces nil

asciilifeform: btw mircea_popescu & diana_coman , not only 0 packet losses, but 0 reorders.

asciilifeform: but ftr i released complete kit with orig v, not half thing.

asciilifeform: i dun recall asking for 'use THIS .emacs' lol

asciilifeform: and when esthlos releases, will try his.

asciilifeform: mircea_popescu: just nao -- muscle-powered v a la diana_coman . this weekend would like to reword v.py to run on phf's components.

asciilifeform: packets, i mean

asciilifeform: mircea_popescu: seems like 100% passed...

asciilifeform: mircea_popescu, diana_coman : http://nosuchlabs.com/pub/udpism/usa_sender_udp_log.txt http://nosuchlabs.com/pub/udpism/uy_receiver_udp_log.txt << 1 full volley ☟︎

asciilifeform: 'Note that the sender will send each size of package *only once* and it will simply finish once it sent one package of each size' << aaah

asciilifeform: oh hm it stops after a while

asciilifeform: diana_coman's test jig ( i did not modify it except for the dest ip ) currently fires 1 / sec.

asciilifeform: going by the current empirical test, however, a packet that frags into 2 or even 3, typically goes. tho it remains to be seen whether they start falling down once you saturate.

asciilifeform: tho as i understand it, they did not account for the 8 byte udp header size, and thereby still fragged.

asciilifeform: i can see the logic, ethernet frame is 1500 , ip header -- 20 byte

asciilifeform: admittedly this was in the paleolithic '90s

asciilifeform discussed subj with asciilifeform's brother, who answered 'whaddayamean, what size packet, at $defunctgamesco we only ever used 1480, for decade, ideal' ☟︎

asciilifeform: interestingly, 1st coupla min seems to show ~0 loss

asciilifeform: will leave it overnight , then post log..

asciilifeform: diana_coman: built & emplaced your sender-receiver, it is running nao, asciilifeformistan <--> BingoBoingostan