#ls2 (irc2p) | #ls2 I2P Protocol Development Discussion - Next Meeting Monday, October 9th 6:30 UTC

eyedeekay1 go-i2p already has an i2ptunnel(that it can't use yet) called "sam-forwarder" which also sort of forms the basis of a bunch of other applications

eyedeekay1 it's the thing that publicly lists the webIRC client in BRB if you choose to share it, it's the thing that forwards the torrent tracker to I2P in samtracker, eephttpd which is... a long story... but anything you would use i2ptunnel for sam-forwarder can be used for too and I use it a lot.

eyedeekay1 And as I said, this forms the basis of something of an ecosystem of applications, one which I've been badly in the weeds on.

eyedeekay1 So an update to sam-forwarder(which resulted in performance and reliability improvements) became the starting point for digging myself out of those weeds

eyedeekay1 sam-forwarder got 2 releases this week, which also entails updating the family tree of software that uses itDuring this process, I'm taking the opportunity to make lists of the software related to sam-forwarder and actually explain how each piece relates a little better, and identify which ones I want to maintain and which ones I want to retire(in a public status update)

eyedeekay1 It's also an opportunity to clean up the absolute mess that is my web site and lessen the extent to which my existing projects weigh me down, and when go-i2p is ready, sam-forwarder and the rest of the Go ecosystem will already be there for it to use

eyedeekay1 I didn't get back into transports until yesterday morning unfortunately, where in order to test I'm just manually constructing routerInfos and routerIdentities instead of trying to read them from a stream or a file or something

eyedeekay1 I have noise-only listeners and connections being established in this way on the same machine

eyedeekay1 So I'm going to write the godoc and check that in today or tomorrow

eyedeekay1 Then I'm on track to try and add the pre-processing and post-processing steps for NTCP2 obfuscation, which I think I can do right on top of

eyedeekay1 the noise transport instead of doing it all in a different struct, which would be very nice if I'm right

eyedeekay1 EOT go-i2p

zzz so you have handshake working in noise (without i2p mods) ?

eyedeekay1 Yeah noise-only, completely unmodified

eyedeekay1 Pretty unimpressive but it's a start

zzz great

zzz anything else on 2) go-i2p ?

eyedeekay1 Nothing else

zzz 3) ack immediate flag for ssu2

zzz so I've been teasing this for a couple of months

zzz that we need something like this, like we have in streaming

zzz to ask for an "immediate" ack

zzz in quotes because our implementation is a delay, but very short, a few ms

zzz so I finally coded and tested it

zzz compared to without

zzz and on testnet with a 20ms RTT, I get a 34% speedup

orignal nad?

orignal do you have some flag like in streaming or what?

zzz with a 50ms RTT, only 5% speedup, within the margin of error

zzz in streaming it's delayrequested=0

zzz basically you set it when your send window is almost full

orignal so what will be in SSU2

orignal ?

orignal because I always ack immediate now

zzz I picked a bit in the flags part of the header

orignal ofc not after every single packet

zzz header[13] & 0x01

orignal good

zzz we have 24 unused bits of flags

orignal I know

orignal but please tell me

zzz seems like a waste to make it a block, that would be 3 bytes

zzz flags bit is "free"

orignal why do you even need to send ack with delay?

zzz delayed ack is much more efficient, only ack once for a burst of packets

orignal I do it without delay

orignal e.g. one ack for burst of packets

orignal it makes sense for streaming

orignal but for tranports I doubt

zzz my code doesn't know about "bursts", it just waits a little while before sending an ack

orignal because it's basically fucking TCP_QUICKACK

orignal I guess I can ignore this flag for now

orignal for receiving

orignal and set it for sending

orignal in last packet

orignal right?

zzz sure, acks are much "cheaper" in SSU2 than in streaming

orignal did I tell about TCP_QUICKACK?

zzz but you still don't want to send 64 acks after receiving 64 packets

zzz sure

orignal as I said I send only one

zzz in streaming we send it for every packet after the send window is >= 2/3 full

zzz so that's the way I did it for now in ssu2 for the test

orignal while TCP_QUICKACK causes major problems

zzz yeah I think you talked about quickack before

orignal my point is, we should start away from it

orignal to avoid future problens

zzz ok

orignal anyway I will start setting this flag for last packet I send in one batch

orignal also another thing

zzz ssu2 is just like streaming, in that the receiver doesn't "know" what the sender's send window is. so the immediate ack flag helps fix this

orignal what if you have thousands of SSU2 sessions

orignal you will have to set this timer for each of thme

orignal how many file descriptiors would you need?

zzz sure, but only a small % have a delayed ack at any one time

zzz timers don't use file descriptors in java

orignal how?

orignal please explain

orignal you receive a single packet

orignal you set timer

orignal and when it expires you send ack

zzz right

orignal and you do it for every receiving

orignal for me it's huge overhead

zzz my median idle time (since last received packet) is 37 seconds

zzz my median timer delay is 12 ms

orignal we can't rlay on it

orignal we should assume contrnious stream of data

zzz don't you already have a per-session retransmission timer?

orignal I would say 10 packets per millisecond

orignal no

orignal I use common resend timer for all session

zzz ok

orignal I check if it's time for each session and do it

orignal it's like every 200 ms of so

zzz anyway, my recommendation is that we add support for an ack-immediate flag. how to implement efficiently is up to you. I've found it provides a substantial speedup

orignal and I can reduce up to 10 ms if necessary

orignal as I said I will do for sending to let you know if I'm done with sending

zzz ok great. I'll add it to the spec and let you know when it's updated, thanks

orignal fine

zzz anything else on 3) ?

orignal no

zzz anything else for the meeting ?

orignal no

zzz that's it then, thanks everybody

zzz eyedeekay, what do you have for language stats?

zzz and what are "Windows" stats?

eyedeekay I don't have windows stats, just a report on when/what/why of the release

eyedeekay But it's pretty much all in the blog

zzz blog? why a blog?

eyedeekay I did a release blog for the easy-install bundle like always

eyedeekay News too

zzz where?

zzz I'm lost

eyedeekay Onhttps://geti2p.net/en/blog/post/2022/08/28/easy_install_bundle_1.9.0

eyedeekay And in the news su3 for the Windows easy-install update

zzz I was looking for user counts by language, extracted from eche's news server, to send to the loc. lab... is that what we're talking about?

eyedeekay I had 3 notes sections in my notes about release this week, Android, Windows, and Language Stats

eyedeekay Right now I'm talking about Windows, language stats was a separate section

zzz I don't see any stats in that blog post

zzz and I don't know why we would put stats in a blog post

eyedeekay I didn't put the language stats in the blog post, the blog post is just about Windows

zzz so I'm even more lost

zzz do you have language stats?

eyedeekay Yes I started collating them this morning when we reached 70k requests since the last update

zzz it's in a separate section of the blog?

eyedeekay No just my notes, I usually come to the meeting with 50 or so lines of typewritten notes on what I've done this week

zzz ok. becoming less lost :)

eyedeekay Just to make sure I don't miss anything

zzz so where might you have the stats?

eyedeekay Well I can share them here or make a private repo and gitlab and give you access

eyedeekay Or just make it public, depending on how we want to handle it

zzz what is it just a CSV or something

zzz ?

eyedeekay Right now it's logs that I've grepped for supported languages in, separated into text files by language, and counted with nl

zzz why don't you just email it to me and we'll go from there

eyedeekay No CSV yet, I started it this morning when request count hit 70k, which is about the average network size reported on stats

zzz it's both languages and countries, right?

eyedeekay yeah

zzz we may need to merge the countries back to just languages

zzz and I want to add a column for translation %

zzz somehow

zzz so we can make some sort of score of most-needed translations

eyedeekay I'm going to analyze totals and proportions

zzz like for spanish, we don't need 52 rows one for each country, i don't think

eyedeekay I'll make it digestable

zzz ok, but there may be some languages where country is important too. not sure how to handle. spend some time on it if you want, or just send me what you have if you don't

eyedeekay I'm going to try and make it a useful overview of the languages and spend an hour or two on it, I'll email that to you tonight

zzz super, thanks

zzz spec updated with immediate-ack flag info, please review