***: nesta_ has quit IRC (Quit: zzz)
nesta has joined #arpnetworks
mnathani has joined #arpnetworks mercutio: hazardous: that's pretty nasty
but clever at the same time.. :)
when i go to the supermarket there is a savings card that remembers/tracks everything you buy
it's kind of handy as it'll tell you when things you frequently buy are on sale.
but some people really don't like that level of tracking. at least it's obvious. fIorz_: ... except it isn't actually obvious? mercutio: it's obvious to me? :)
i suppose it isn't necessarily obvious to everyone staticsafe: https://patchwork.ozlabs.org/patch/671069/ that is a serious patch, damn mercutio: seems short to me
it's a new congestion algoritham
oh serious as in great
not serious as in big
it requires fq
wow
i want to try it now staticsafe: heh mercutio: did you try it static? staticsafe: not yet fIorz_: mercutio: well, yeah, it's probably obvious to anyone in here--but I think for most people, it's not obvious at all, and in particular what the total consequences are or could be, it's rather just seen as some nice gift of the supermarket mercutio: i played with heaps of congestion stuff years back
i found bic can sometimes be faster than cubic
and i found that the new initcwnd stuff from google really helped
but that 8 packets worked better than 10 sometimes
and that 12 worked better than 10 sometimes
10 was always an in between amount. but i think they went too high seriously.
also i found that pacing packets doesn't help nearly as much as i thought it would.
in normal internet packets come in bursts, on/off. but that means recovery works better too
but there's some empty space on the off
there's a sysetm to send less packets in the off time
which works better than trying to haev an average send speed.
also waking up to send packets frequently doesn't work well with regards to timers etc.. it puts cpu load up
atm linux has some serious issues with granuality with tc
and the shaping actually doesn't scale nearly as well as it should because of timer issues. staticsafe: :o mercutio: you need really really tight timers to do shaping well BryceBot: That's what she said!! mercutio: like more than 1000/sec
esp if you can't have more than 2 or 3 packets in flight at once
but 1000/sec with 3 packets in flight is what 3000 pps
so when you want 30k+ pps per queue you start seeing the granuality needs to go up fIorz_: that BBR patch indeed looks nice mercutio: hmm that's using estimated bandwidth
that doesn't work on lots of current networks :(
becuase of issues like that shaping/pacing issue
people will have gigabit connects rate limited to 200 megabit
so traffic estimates can get confused
cos rate limit is like "go over threshold, packets dropped"
end site queueing is where the most dramatic improvements can happen atm i reckon
i have dsl anyway, traffic estimating should work ok :) staticsafe: i wonder how big of a problem asymmetric end-user connections are mercutio: not at all
unless one side is too small
lack of decent aqm/queues are a way bigger problem
like if your connection receives 30 packets back to back will all 30 get through
and what will happen to latency
with google's 10 packet initcwnd thing being common now that's just 3 web responses simultaneously
before even increasing window size
14600 bytes * 3
43.8K, then acceptable latency spike
say you want 5 msec
you'd have to have 200*43.8k/sec throughput
which means 8760k/sec, so about 80 megabit
so as soon as you have 8 megabit connectoin
latency spikes to 50msec...
which is definitely too high to my midn. so 10 packet initcwnd is too high for 8 megabit
but aqm/shaping at the end point can help
like fq_codel, which that algoritham relies on heh
but want fq_codel at end point too staticsafe: hm what is Mikrotik RouterOS using my default mercutio: wifi is actuall even worse staticsafe: never looked into it mercutio: hardware queues
you need it before it hits you though
but for outbound you can set sfq
i don't think they have fq_codel yet
i don't actually have any outbound aqm atm
only inbound
i don't really do that much uploading, and can do a bit over 30 megabit up now
on 1 megabit you really need aqm though staticsafe: only-hardware-queue leaves interface with only hw transmit descriptor ring buffer which acts as a queue in itself. Usually at least 100 packets can be queued for transmit in transmit descriptor ring buffer. Transmit descriptor ring buffer size and the amount of packets that can be queued in it varies for different types of ethernet MACs. mercutio: what connection type do you have static
think i'm going to test this staticsafe: mercutio: DOCSIS3 cable
250/10 mercutio: yeh you mostly haev to care up probabluy staticsafe: indeed mercutio: you can do a 10 megabit queue on your outbound with burst
then put sfq on top of it
it's probably slightly over 10 megabit
so 10 megabit should work ok
fq_codel would be nicer though
Hunk #1 FAILED at 124.
nearly applied cleanly
actaually i think that's fine, it's only diag stuff
i wish vultr had serial console :)
a working console would be a good start staticsafe: heh mercutio: hard reboot fixed it...
damn vultr sucks :)
then kernel.org mirror is down
there's also a new updated vegas algorirtham now
hmm need linux/win_minmax.h
oh
i think we need all 14 of those patches
hmm 16. does anyone konw an easy way to patch all 16?
manually clickign seemed to be the easiest as i couldn't find mbox of month
# cat /proc/sys/net/ipv4/tcp_available_congestion_control
cubic reno bbr cdg nv
hmm it's slower than cubic for peak performance, and faster than cubic for minimum performance
and on large file much more consistent, and 23 seconds vs 26 seconds staticsafe: how are you testing? mercutio: with curl
from vultr sydney
to dsl end point
i wanted something closeish, but not too close
i'm getting a 200mb file
i was doing 10mb at first though
the smoothness is way better with bbr
like while watching the curl current speed
where cubic is going up and down like a yoyo
so it's 8688k/sec with bbr, and 7855k/sec with cubic
that's basically 10% faster. this is with 85 megabit sync rate on vdsl
i should test from further away too
19.2/18.9 to sj
16.0/15.8 to arp
so yeah it looks like it should in general give some bandwidth increase
even if only a couple of megabit mnathani: is this some low level tcp/ip enhancement mercutio: it's a new congestion control algoritham mnathani: thought we only needed those when using legacy network gear like frame relay mercutio: this will make the internet go faster in general mnathani: s/algoritham/algorithm BryceBot: <mercutio> it's a new congestion control algorithm mercutio: but it's really wifi/3g/4g etc that are the problem areas atm mnathani: faster is better mercutio: as latency and bandwidth can vary dramatically over time mnathani: is it Linux only at the moment? mercutio: yeah
it's not even in linux main
it's a series of 16 patches to apply mnathani: Windows can implement it when it gets popular? mercutio: you can try my kernel if you have linux :)
i'm using 4.8-rc6
as it seemed best to use current with it
windows doesn't really matter
it's about the tcp sender
and windows shouldn't be on servers. :) mnathani: I am on Linux ixp 4.4.0-21-generic #37-Ubuntu SMP Mon Apr 18 18:33:37 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux mercutio: https://weallsee.net/vmlinuz-4.8-rc6
it should have all the important things not in modules :)
and only suitable for virtual machines mnathani: ofcourse I should test before applying that mercutio: ie it doesn't have heaps of hardware support for weird stuff mnathani: to see if it improves mercutio: nah
you can change real time
echo bbr > /proc/sys/net/ipv4/tcp_congestion_control mnathani: change kernel withough reboot? mercutio: and echo cubic ..
yeah
oh
nah you can change congestion algoritham mnathani: ahh that makes more sense mercutio: i just have alternate kernel, you could compile yourself if you want
i can tar up the patches mnathani: as long as it works with ubuntu I dont mind trying your kernel mercutio: cool
yeah i'm using it with ubuntu
just stick it in /boot
and run update-grub mnathani: how do I set the default kernel
or will it do that automatically mercutio: update-grub will set the highest number first
you'll see it when it runs update-grub ***: nathani has quit IRC (Quit: WeeChat 1.4) mercutio: Found linux image: /boot/vmlinuz-4.8-rc6
Found linux image: /boot/vmlinuz-4.4.0-9136-generic mnathani: rebooting now mercutio: like mine is like that
did it come up fine? mnathani: dont think so
I cant ping it mercutio: oh, weird
vnc it? mnathani: yea
console is black mercutio: oh hmm
i think console is normalyl black actually mnathani: enter doesnt seem to change it
not accepting keyboard input mercutio: serial it? :) mnathani: doesnt the OS have to have something configured for that to work
never used serial console before mercutio: ahh
yeh console=ttyS..
console=ttyS0, etc
i wonder why it doesn't show anything on console
and why network isn't working
what file system are you using? mnathani: booting again
kernel panic
not syncing mercutio: remove quiet and splash?
that could be fs loading issue mnathani: unable to mount root fs on unknown block mercutio: oh i think i know the issue
hangon
it's the scsi modules mnathani: I should boot into old kernel? mercutio: hmm
if you want
it has virtio-scsi
gah
does it say partition talbe mnathani: I mean to load a new patched kernel you provide :-) mercutio: like sda1, sda2 etc
well i'm not sure why it's not working atm
can you show me screenshot? mnathani: Device Boot Start End Sectors Size Id Type mercutio: i mean when linux boots is it seeing that mnathani: https://gist.github.com/mnathani/01a7492abc39781e5faa1fd63241d51e
I switched back to the original kernel mercutio: oh you're using lvm
i wonder if i'm missing lvm support
ok compiling again
hmm
i wonder if cloudflare will cache mnathani: you could name it something different mercutio: i disabled caching on cloudflare temp
development mode
it's nearly finished compiling.. mnathani: have you heard of bonded internet: http://itel.com/services/bonded-internet/ mercutio: ok
new version is there
md5 is a39815975d919495da9362a2173e7add mnathani: same filename? mercutio: nah hadn't heard of it
we're getting gigabit internet in this country anyway. mnathani: there is a video as well
https://www.youtube.com/watch?v=tqbnjgbtDl0 BryceBot: YouTube video: "Double or Triple Your Internet Speed - This Method Actually Works!" by LinusTechTips mercutio: it's kind of crazy considering we were far behind mnathani: says you get things like static IP
and unrestricted ports etc
using residential links and their service
looked pretty cool to me mercutio: mptcp is interesting from the pov of things like 4g and home internet
and going from one to the other mnathani: same thing
http://imgur.com/a/bMpgo mercutio: was the md5ok? mnathani: yea
only compared the last 4 characters mercutio: yeah i understand BryceBot: That's what she said!! mercutio: shoudl be fine :)
it really doesn't show much on that screen
i wonder if you could show what it shows on serial and set serial?
is the lvm encryped? mnathani: can I pass a parameter from grup mercutio: encrypted? mnathani: dont think so mercutio: yeah
console=ttyS0,115200 mnathani: on which line
I have setparams ... BryceBot: That's what she said!! mnathani: BryceBot: no BryceBot: Oh, okay... I'm sorry. 'I have setparams ...' mercutio: did you login to serial console? mnathani: there is a linux line
I am on web vnc
grub bootloader mercutio: you have to login to serial console for serial console capturing :)
like ssh to console.cust.arpnetworks.com mnathani: where I can edit boot options mercutio: otherwise won't be able to see it mnathani: ok
but doesnt the ubuntu have to have that command as well? mercutio: what do you mean?
you setill have to set in grub..
but you have to already be on serial console
otherwise you miss mnathani: grub permanent mercutio: there's no storing of previous serial console output mnathani: or grub one time while booting mercutio: once you are logged onto serial console and see the boot coming up etc
you can then set console=ttyS0,115200 as advanced option to grub
err to linux
you can also update your grub config to always do both
GRUB_CMDLINE_LINUX="console=ttyS0,115200n8 console=tty0" ***: Nahual has joined #arpnetworks
Nahual has quit IRC (Quit: Leaving.)
nathani has joined #arpnetworks mnathani: http://www.extremetech.com/computing/234881-from-this-point-forward-all-intel-and-amd-cpus-are-windows-10-only
sounds a bit odd, CPUs not supporting older operating systems mike-burns: The phrasing seems weird. It seems more like MS is not supporting new processors on the older OSes? BryceBot: Seriously guys, are we not doing phrasing anymore? mike-burns: OK, it's just the headline that is oddly phrased. mercutio: wasn't that going around ages ago mike-burns: It's over two weeks old. mercutio: even longer than that i was thinking
jan 16th i found osmething mike-burns: If that's ancient, I'd hate to know what you think of software that was written before then! ***: nathani has quit IRC (Quit: WeeChat 1.4) mercutio: well 8 months is quite a while
http://www.theverge.com/2016/1/16/10780876/microsoft-windows-support-policy-new-processors-skylake BryceBot: The Verge: "Microsoft says new processors will only work with Windows 10" mike-burns: That headline is also oddly phrased! mercutio: heh
trying out this new bbr algoritham, reminds me again of how far behind openbsd is
More details about BBR will be published in the ACM Queue, Vol. 14 No. 5, September-October 2016, under an article titled "BBR: Congestion-Based Congestion Control."
i wonder if that's free mnathani: Access not allowed, your library account has expired
:-)
my University access would have had it mercutio: hmm
i don't think it's actually out yet
http://queue.acm.org/issuedetail.cfm?issue=2984629
ahh it does cost
how do you find performance compares mnathani ?
https://www.kiwicon.org/blue
kind of eerie