#ls2 (irc2p) | #ls2 proposal review channel : No reviews currently scheduled

zzz what does "high performance" mean if it's different from bandwidth?

R4SAS Happy new year, guys!

orignal bandwidth means your network is wide

orignal performance means your device is powerfull

orignal say RPi can sit on high bandwidth but low performance

orignal the idea of this flag tell that high preformance router is capable enough to handle serious load

orignal in this case high loaded tunnels should be built through such high performance routers

zzz good morning and happy new year

zzz how would we measure "performance" and what would the threshold be for "high"?

orignal you too

orignal in first attempt it's simple

orignal on x64

orignal or you specify it explicitly

orignal basically my problem is that "X" is not always "X"

orignal e.g. "X" routers drop packets

orignal also I think it would be nice to have one more code "Q"

orignal like "up to 100 Mbs"

zzz thing is

zzz I don't think we can ever trust caps and use it to send that router lots of tunnels

zzz sounds insecure

zzz have to do it with profiling

orignal my point is opposite

orignal if you don't want your router get flooded don't set it as "high performance"

orignal if you believe you are capable to do it, set it

orignal if you are on arm it's always off

zzz so then we need a "low performance" flag

orignal a meesage to ddoser ))

orignal X was good as "high performance"

orignal until now

orignal Ilita transfers 100 Mbs now but publishes X

orignal limited by port there

zzz we don't do anything with any of the caps really

zzz we don't trust any of it

zzz so not much use for Q

dr|z3d what orignal appears to be referring to is what we have already in java. isSlow()

orignal my point is that X is not really unlimited

dr|z3d I'd suggest that the router itself should determine its performance level, not other routers.

orignal how do you know is your peer is slow?

orignal yes, that's my proposel

dr|z3d you don't. you limit your own router.

orignal to let router tell other if it's powerfull enough

dr|z3d if a router is configured as isSlow() then it handles less tunnel requests, throttles sooner etc.

orignal basically I would like to build high loaded tunnels trough high perfarmance routers

orignal yes, and it should be use for expolratory or IRC only

dr|z3d just keep an eye on the cpu load on the router, and start declining requests when it gets too high. don't need an explicit cap for that.

orignal the worst thig is

orignal a routers accepts a tunnel

orignal and then start dropping messages

orignal I prefer to know if a router I'm building a tunnl trough is capable enough

orignal before I send build request

dr|z3d sure, not great. so the router itself should be aware of the percentage of messages it's dropping, and if necessary self-throttle.

zzz problem is, a cap doesn't help you know if that router has capacity for one more tunnel *right now*

orignal I don't care if a router set code 30

orignal I care if it starts dropping

dr|z3d that's why I don't think a cap is the best idea. you want the router itself to be aware of what its limits are.

zzz I could be medium capacity but have no load now

zzz I could be high capacity but under extreme load now

orignal the problem is that profiling finds a rouuter good on 100 Kbs

orignal and it's out of limites on 1 Mbs

dr|z3d publishing a high perf cap would just be an invitation to route traffic to high perf routers, and then they're overloaded.

orignal dr|z3d yes and our routers are fine with it

orignal R4SAS has 150 Mbs on his dedic for example

orignal and can handle much more

orignal and you don't need high performance routers for IRC tunnels

zzz maybe F = medium loaded, don't use me so much; G = highly loaded, Go away

dr|z3d I don't think this is going to be resolved by publishing a high perf cap. It's going to be solved by the router itself setting limits based on current performance.

dr|z3d yeah, and I was seeing 42MB/s on a router before we started blocking offenders.

dr|z3d doesn't much matter what router you use for irc, it's not going to be demanding much of the router.

dr|z3d except not F.

dr|z3d f is floodfill, so we don't want F. just confuses things.

dr|z3d not even sure we need to publish anything. I still think the router itself should accept or reject requests based on self-profiling metrics.

zzz D=Delayed (medium congestion) E=Empty(high congestion)

orignal yes good point

orignal masyb even separate optional property?

zzz you could do capacity=75%, or tunnelsAvailable=123, or bwAvailable=3MBps, but I'm not sure what other routers would do with that

zzz how does that minutes or hours old info help you decide whether to route 1 tunnel through him?

orignal if it's one hour old

orignal it might be right after connection

dr|z3d no one's telling me why more self-awareness for routers isn't the best solution to this perceived problem.

dr|z3d I don't need to tell you what I think my current capacity to host your tunnels is, I just don't host them if I'm under load.

dr|z3d router to router, I might classify my tunnel requests as high bandwidth, or low latency, or perhaps both, with some requirements, and your router can accept or deny requests based on current performance. there was a paper published a while back that suggested difference classes of tunnels.

dr|z3d the problem with setting performance thresholds for tunnels is that very soon everyone wants the fastest, lowest latency tunnels and all the network traffic gets funneled to a subset of routers.

orignal hmm

orignal 4rum.i2p/topic/7

dr|z3d one thing you can do off the bat, orignal, is reduce the default max transit tunnels for arm/32bit and whatever else you deem to be slow.

orignal for me number of trsit tunnels is not an issue

orignal only traffic

dr|z3d orignal: I use the isSlow() category to limit transit tunnels, max bandwidth share, min transit tunnels before throttle etc. combined, they should collectively serve to keep underpowered routers from handling more traffic than they're capable of, unless a user overrides the defaults.