***: grepidemic has joined #arpnetworks
m0unds has quit IRC (Quit: derf)
m0unds has joined #arpnetworks
novae has quit IRC (Ping timeout: 245 seconds)
carvite has quit IRC (Remote host closed the connection)
Erick- has joined #arpnetworks Erick-: !help brycec: 89With?
*With? ***: carvite has joined #arpnetworks
m0unds has quit IRC (Quit: yerf)
mus1cb0x has joined #arpnetworks mus1cb0x: anyone know why i'd be getting silent broken pipes to my vps on kvr14 lately? mnathani: mus1cb0x: what OS? mus1cb0x: fbsd 9.2
same as another vps on another kvr which doesn't break pipe ***: m0unds has joined #arpnetworks mnathani: Is there such a thing as a development version of the internet, a place where one can test BGP configurations without the risk of messing up any production networks? m0unds: labs
build a lab w/a few devices that speak bgp
then you can simulate to your heart's content mercutio: mnathani: it's called a local network
if you want a full route table you can probably get one passively that you can't route
you can't really mess up much with bgp as long as both ends have filters.
so if the other end filters what ip addreses you can adverrtise, you just set aside those ip's to advertise and make sure they're not in use from other things
then advertise those ip's...
and you can use a private asn if you don't have a bgp asn mnathani: Definitely need to start with a private ASN
does IPv6 work pretty much the same as v4 with BGP? mercutio: you can reuse an asn in more than one place on a network
yeah
you have ipv4 and ipv6 bgp sessions separate
and ipv4 is over ipv4 and ipv6 is over ipv6
subnetting is more complicated :)
cos so many ip's to choose for link address
but lots of people are using /64s to connect to each other on ipv6
and /30s on ipv4
in theory you can use /31s and /127s mnathani: > /48 per site mercutio: nah for the link address to the bgp neighbour
for directly connected links mnathani: right mercutio: i wonder what arp does mnathani: a /48 is like 65000 /64s right? mercutio: a /64 can lead to neighbour discovery
err implies
which isn't really necessary for connecting point to point links mnathani: Event if you do use a /127 you are not likely to use any other subnets from that /64 for anything else? ***: novae has joined #arpnetworks mercutio: blah blah power problems m0unds: POWAR mnathani: How would I test packet prioritization on a local LAN? mercutio: use iperf to blast udp traffic? mnathani: if voip traffic was being prioritized - the iperf traffic should transmit with a lower priority / dropped packets? mercutio: yeh
where's it being prioritised?
if you go gigabit -> 100 megabit -> 10 megabit mnathani: presently in a mikrotik vm mercutio: wehre 100 to 10 is the itnernet and gigabit to 100 megabit is the router
then flooding 100 megabit will still give you issues
but you should be able to flood 20 megabit and have a voice call
usually fast to slow connectinos don't work very well without aqm mnathani: I guess I would flood with the prioritization disabled to compare mercutio: and switches don't normally have good aqm
if you're going gigabit to gigabit to 100 megabit
then linux can congest the queue on the ethernet
reducing txqueuelen or using aqm can help
fq_codel is kind of cool and easy to configure, and works fine at gigabit speeds
tc qdisc add dev eth0 root fq_codel
only in recent kernels though
on older stuff i would normally just use sfq
but sfq doesn't work well for things like bittorrent, but it's generally fine for light networks
ie mixing buik/interactive
sfq is basicallyl like having 8 queues
and randomly splitting between them
so it usually works
but if you have lots of bulk senders it doesn't tend to work
it may not be as bad as without, but won't really fix things
normal linux has some prioritisation too
but you have to reduce queue length to really benefit
and uhh change ring size on the ethernet driver m0unds: w/mikrotik it's easier to use a queue tree w/global queues and subqueues
makes it easier to visualize the traffic too
as far as impact of qos on it, ensuring stuff you expect to be hitting a particular queue is in fact hitting it and being impacted the way you intend mnathani: how does the 'priority' field work, presently my queue has it set to 8
is higher better, or lower?
and the queue types confuses me: pfifo vs sfq, vs pcq, vs red mercutio: i wouldn't even touch priorities myself
just use aqm
voip is pretty tolerant
skype is worse
because you tend to talk to people further away,.. m0unds: http://wiki.mikrotik.com/wiki/Manual:Queues_-_PCQ mercutio: you can only relaly fix outbound traffic too m0unds: ehh, you can apply downstream limiters to prevent latency from spiking mnathani: can you do something like allow download upto and untill ping to certain destination gets higher than desired limit, kind of variabla like that? m0unds: i just tested with my particular connection
pull down a ton of traffic, note when latency starts getting out of hand, limit to a point before that
same with upstream
with my 50/10 at home, it ends up being 48 down 8 up mnathani: so once those limits are set, pretty much no amount of downloading would drastically increase latency? m0unds: right
i could saturate my downstream at home to 48mbit without latency and jitter getting wacky
so i rate limit traffic to 48 down, 8 up and then assign classes to traffic types to give higher priority to certain traffic
the higher priority part allows multiple high-bandwidth applications to operate on my network without degrading voice or gaming or whatever
my goal was to make it possible for me to play $somegame with my wife streaming HD video from amazon or netflix without impacting gameplay
was trying to see if i had a copy of my mikrotik config somewhere as an example. seems like i don't mnathani: what mikrotik model do you use? m0unds: i had an rb450g for a long time, it was an early hw rev and had some power circuitry issues. when it died, i swapped it for an rb1100ahx2 because i got a good deal on it ($200) and ended up selling that to buy an SRX210he because the 1100 was way too quirky for me ***: novae has quit IRC (Ping timeout: 255 seconds)
novae has joined #arpnetworks
grepidemic has quit IRC (Quit: leaving)
grepidemic has joined #arpnetworks
acf_ has quit IRC (Read error: No route to host)
grepidemic has left "WeeChat 0.3.8"
acf_ has joined #arpnetworks
grepidemic has joined #arpnetworks
mus1cb0x has left "WeeChat 0.4.2" mercutio: mnathani: fq_codel is meant to behave kind of like that
http://www.bufferbloat.net/projects/codel/wiki