woot, got a 10x improvement in guest to host network performance
http://man7.org/linux/man-pages/man7/lvmcache.7.html
has anyone seen this before?
acf_: i am trying to access your smokeping.  did you disable it for external access?
hmmm..
also your / exposes more than it should, but loads.
seems there's some congestion tonight
yeh
on level3
can you do a mtr?
yea, it's the return path
i'm trying to track it down, but i reckon it's level3
it is
it's really bad to my uk server.
it's ugly
haha
i'm seeing 60ms +/- 150
oh i just saw loss, and throughput issues
udp loss was higher than icmp loss
but latency seemed normal
but throughput seemed even worse than the udp loss
i just got a vultr vm that's on level3
there's only one location in the US that routes via level3 from arp, most are ntt
yeah, udp loss is gross
what's your ip?
like 50+%
arp IP?
oh wow i didn't see it that bad before
nah ip you're coming from
73.26.90.138
http://pastebin.com/HUZS2VD0
yeah it loosk strange
not as bad that run
it's up and down a lot
yeah
that's with -u?
yeah
on the bottom, forgot to redo the top
yeah do icmp
it's bloody hard to read with all that hop alternating
it seems you're not getting loss just latency from that though
and from my mtr from arp to you it's the same
it actually looks like comcast issue in this instance :/
looked largely the same
yeah just trying to track down wehre it's happening
try doing 500 of them
i realise it takes nearly 10 minutes :)
but if there's a couple of percent loss or something it shows it up better
well, there's nearly always loss at any time of the day on level3 w/icmp
yeah on those earlier hops
that's been my experience w/them forever
it's whether it carries through
i don't tend to have any loss at the destination, just along the path
but ssh is laggy?
but when i mentioned congestion, i saw a lot of loss. it seems to just be high latency now with no loss
it isn't
That's what she said!!
BryceBot: no
Oh, okay... I'm sorry. 'it isn't'
i'm getting good speeds from this level3 connecting vultr vm :(
ahh ok
ssh wasn't laggy it seemed to uk for my "cl" program
but i was getting really bad throughput
starting around 6 hours ago
i updated my kernels and got some speed boost in general
http://uk.meh.net.nz/cgi-bin/smokeping.cgi?target=Curl.chi
i should fix the naming
that's la.meh.net.nz
http://uk.meh.net.nz/cgi-bin/smokeping.cgi?target=meh.lameh
but ping has been ok
and loss isn't that high
http://pastebin.com/BNiUsqcP
comcast's routing is changing a lot grr :/
what throughput do you get atm?
it seems it's going via los angeles or denver
yea, it changes depending on time of day
sometimes it's direct abq > lax
it changes within the trace
sometimes it's denver > lax
but that may be the jitter
route change adds 10ms to ping but that's it
throughput is nearly always the same
and it's still ok?
yep
hmm damnit
so that's different to the other issue :/
actually nope
it just tanked
was going 1.2MB/sec, dropped to 400KB/sec
that's pretty slow both times
yeah that's more like wha ti been seeing
except lower because it's from uk
well, here
higher ping makes loss worse
http://pastebin.com/3HhgkHDa
two local hops from arp > comcast are gross
http://pastebin.com/9FhM7qHX
yeah, that same 150-200ms "worst" measure is in the same spot as mine
5 & 6
yaeh hard to know if that's deprirotisiation or not
the latency to hop 5 for me is going hmm
bi'm still running my download heh
average speed 49305 bytes, current speed 45476
it was going 80k/sec before
anyway why's it fine with this vultr vm grr
by fine i mean 7.6mb/sec
oh shit
?
it's coming in from vultr via nlayer
i wonder if that has any bearing
oh and now it's showing loss on last hop
and on mzima pings heh
did you stop your dl?
nope
dl slowded down
its' getting slower nad slower
funky
27k/sec now
wow
yeah i know it's weirdly slow
and how does 6% icmp loss fit in with that?
with 7.6mb/sec
damnit i have to wait for dns ttl :(
this is making no sense
so there's packet loss from la to miami
but throughput is good
and the packet loss starts at gtt in san jose
and throughput is good in both directions
maye it's bursty
heh " curl -q http://weallsee.net/testspeed | sh
"
is convenient
but it feels dirty just running stuff downloaded off the internet
it could be anyhting
m0unds: i'm setting high jitter to you from jp.meh.net.nz over ntt too
so it may be your path to arp
or within comcast's netowrk
i'm loading my connection up quite a bit
atm
ahh ok
maybe that's it then
i'm trying to get my systems better organised for pushing smokeping etc :)
mercutio: what os are most of your vultr vps?
ubuntu
one is arch
if there was arch auto provisioning they'd all be arch
i've just shifted them all to custom kernel though
any performance benefits?
a little
to the kernel
i'm not sure why :)
i did the kernel when i was trying to debug guest-host kvm performance at home
Startup finished in 944ms (kernel) + 491ms (userspace) = 1.435s
that's sj
Startup finished in 948ms (kernel) + 732ms (userspace) = 1.681s
and that's my home vm
home vm is ubuntu
and tcp performance went up a little
mercutio: hmm, haven't touched the smokeping config
probably just congestion or something
huh. it's acting funny for me too
[....] Starting latency logger daemon: smokeping
>:(
all of the smokeping things are blocking on something
huh. that was really weird
seems that smokeping was blocking on DNS resolution
both the CGI and the service
damn
i wonder why
oh it didnt' seem to lose data?
cool 4.2.2.1 is screwed
from about 2pm your time
aare you pdt?
hmm and phoenix.acfsys.net
is that a box you can trace from?
oh it's a tunnel or something