[Snap] UFB Poor Performance

Forums › 2degrees (including Slingshot, Orcon, Flip, Stuff Fibre, MyRepublic, 2talk and Vocus) › [Snap] UFB Poor Performance

1 | ... | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | ... | 19

195 posts

Master Geek

#688914 20-Sep-2012 16:05

% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 77.2M 100 77.2M 0 0 675k 0 0:01:57 0:01:57 --:--:-- 862k

From the 149 address I got a much slower download but I started another one just to check afterwards and multiple connections is still maxing the connection pretty much.

Here is the other one

% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 100M 100 100M 0 0 3825k 0 0:00:26 0:00:26 --:--:-- 4100k

insane

3239 posts

Uber Geek

ID Verified

Trusted

#688951 20-Sep-2012 17:09

fellaintga:

...its not Snaps international bandwidth there is a whole lot of factors.

I don't think anyone is insinuating that, more trying to work out why some UFB connections are slower than ADSL2+ or VDSL2 for international downloads, yet still capable of pulling respectful national speeds.

hads

392 posts

Ultimate Geek

Trusted

#688961 20-Sep-2012 17:15

I would imagine it's just the issue that Ragnor posted about back on page 3.

nicegear.co.nz - VoIP, DVB and other Linux compatible hardware

rhysb

435 posts

Ultimate Geek

Trusted

#688964 20-Sep-2012 17:19

Some more from my Sanp 100/50 connection. International does look poor.

wget http://149.20.4.69/pub/linux/kernel/v3.0/linux-3.5.4.tar.bz2
--2012-09-20 17:00:18-- http://149.20.4.69/pub/linux/kernel/v3.0/linux-3.5.4.tar.bz2
Connecting to 149.20.4.69:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 80980207 (77M) [application/x-bzip2] Saving to: `linux-3.5.4.tar.bz2.1'

100%[======================================>] 80,980,207 211K/s in 6m 29s

2012-09-20 17:06:48 (203 KB/s) - `linux-3.5.4.tar.bz2.1' saved [80980207/80980207]

and

wget http://fry.snap.net.nz/testfile-100mb
--2012-09-20 17:10:22-- http://fry.snap.net.nz/testfile-100mb
Resolving fry.snap.net.nz... 111.69.33.8
Connecting to fry.snap.net.nz|111.69.33.8|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 104857600 (100M) [text/plain] Saving to: `testfile-100mb.1'

100%[======================================>] 104,857,600 8.19M/s in 12s

2012-09-20 17:10:35 (8.11 MB/s) - `testfile-100mb.1' saved [104857600/104857600]

Using a multithread download for http://149.20.4.69/pub/linux/kernel/v3.0/linux-3.5.4.tar.bz2 it peaks at about 2MB/s

mercutio

1392 posts

Uber Geek

#688971 20-Sep-2012 17:27

insane:
fellaintga:

...its not Snaps international bandwidth there is a whole lot of factors.

I don't think anyone is insinuating that, more trying to work out why some UFB connections are slower than ADSL2+ or VDSL2 for international downloads, yet still?capable?of pulling respectful national speeds.

i reckon it's however they're being shaped/rate-limited or too small buffers. (both of which are kind of related)

if it's gigabit speeds then suddenly all packets are dropped after 100 megabit, there could be considerable packet loss as speeds go up.

although that doesn't quite explain 200k/sec :)

quakeguy

111 posts

Master Geek

Trusted

#689202 21-Sep-2012 08:26

It's very easy for me to be biased, but are the 'sources' we're talking about, actually $30/month cheapies? You're not going to get much off OVH/Hetzner/Linode with a single thread... I've had boxes at all three and currently operate two at Hetzner and one at Hurricane Electric.

Try my Hetzner box for example: http://de2.tw.net.nz/testfile

If you're using some type of Linux, compare 'wget' or 'curl' to 'axel' - multiple threads seems to help. E.g.

axel -a -n 10 http://de2.tw.net.nz/testfile

That's one of their Falkenstein datacenters. The Nuremberg DCs bandwidth appears to be not as good.

“I do not think there is any thrill that can go through the human heart like that felt by the inventor as he sees some creation of the brain unfolding to success... Such emotions make a man forget food, sleep, friends, love, everything.” - Nikola Tesla

Disclaimer: Views expressed in my posts do not necessarily reflect those views of my employer.

hads

392 posts

Ultimate Geek

Trusted

#689205 21-Sep-2012 08:31

I completely understand that Linode are a cheap VPS hosting provider, but, if I can pull a pretty consistent 1.5MB/s off it with my Telecom ADSL using a single TCP connection then it shows it's not the far end that is the issue.

nicegear.co.nz - VoIP, DVB and other Linux compatible hardware

Equinox IT

Cloud spending continues to surge globally, but most organisations haven’t made the changes necessary to maximise the value and cost-efficiency benefits of their cloud investments. Download the whitepaper From Overspend to Advantage now.

quakeguy

111 posts

Master Geek

Trusted

#689212 21-Sep-2012 08:43

hads: I completely understand that Linode are a cheap VPS hosting provider, but, if I can pull a pretty consistent 1.5MB/s off it with my Telecom ADSL using a single TCP connection then it shows it's not the far end that is the issue.

At certain times of the day I can pull 6MB/sec from the server at Hurricane. But I won't post the URL to that, because the NIC is only 100mbit.

Disclaimer: Views expressed in my posts do not necessarily reflect those views of my employer.

mercutio

1392 posts

Uber Geek

#689396 21-Sep-2012 13:29

quakeguy:
hads: I completely understand that Linode are a cheap VPS hosting provider, but, if I can pull a pretty consistent 1.5MB/s off it with my Telecom ADSL using a single TCP connection then it shows it's not the far end that is the issue.

At certain times of the day I can pull 6MB/sec from the server at Hurricane. But I won't post the URL to that, because the NIC is only 100mbit.

is the germany one not 100 megabit? i got 11mb/sec for a while from amsterdam, then it dropped to 80k/sec. uk was 5mb/sec average, 7mb/sec peak, los angeles 3mb/sec, nz was about 1.5mb/sec.

it also seemed like the host was pausing/stuttering a bit at the start, and raising up in bandwidth a bit slow. maybe it's not using cubic congestion control?

when i first tried vps's i found that openvz network performance was often worse than xen for long distances, partially due to them generally using old linux 2.6.18. since then there have been quite a few improvements that help "nz internet".

generally speaking, vps's often seem to do about 3mb/sec, unless nearby/close where it can be 10 to 20 megabytes/sec. a lot of the bigger ones have 10 gigabit pipes through cogent and/or he.net, plus local peering points which should mean "ok" performance from NZ.

quakeguy

111 posts

Master Geek

Trusted

#689398 21-Sep-2012 13:32

mercutio:
quakeguy:
hads: I completely understand that Linode are a cheap VPS hosting provider, but, if I can pull a pretty consistent 1.5MB/s off it with my Telecom ADSL using a single TCP connection then it shows it's not the far end that is the issue.

At certain times of the day I can pull 6MB/sec from the server at Hurricane. But I won't post the URL to that, because the NIC is only 100mbit.

is the germany one not 100 megabit? i got 11mb/sec for a while from amsterdam, then it dropped to 80k/sec. uk was 5mb/sec average, 7mb/sec peak, los angeles 3mb/sec, nz was about 1.5mb/sec.

it also seemed like the host was pausing/stuttering a bit at the start, and raising up in bandwidth a bit slow. maybe it's not using cubic congestion control?

when i first tried vps's i found that openvz network performance was often worse than xen for long distances, partially due to them generally using old linux 2.6.18. since then there have been quite a few improvements that help "nz internet".

generally speaking, vps's often seem to do about 3mb/sec, unless nearby/close where it can be 10 to 20 megabytes/sec. a lot of the bigger ones have 10 gigabit pipes through cogent and/or he.net, plus local peering points which should mean "ok" performance from NZ.

Yes, the German box is 100mbit/sec. The reason I am happy to post that one, is that it's not doing anything else with that bandwidth :-)

root@de:/var/www# cat /etc/issue
Ubuntu 11.10 \n \l
root@de:/var/www# uname -r
3.0.0-14-generic-pae
root@de:/var/www# cat /proc/sys/net/ipv4/tcp_congestion_control
cubic
root@de:/var/www# cat /proc/sys/net/ipv4/tcp_ecn
2

I'm getting a little stuck for nice machines to test from, so if you have anything, ...

Disclaimer: Views expressed in my posts do not necessarily reflect those views of my employer.

mercutio

1392 posts

Uber Geek

#689401 21-Sep-2012 13:34

quakeguy:
mercutio:
quakeguy:
hads: I completely understand that Linode are a cheap VPS hosting provider, but, if I can pull a pretty consistent 1.5MB/s off it with my Telecom ADSL using a single TCP connection then it shows it's not the far end that is the issue.

At certain times of the day I can pull 6MB/sec from the server at Hurricane. But I won't post the URL to that, because the NIC is only 100mbit.

is the germany one not 100 megabit? i got 11mb/sec for a while from amsterdam, then it dropped to 80k/sec. uk was 5mb/sec average, 7mb/sec peak, los angeles 3mb/sec, nz was about 1.5mb/sec.

it also seemed like the host was pausing/stuttering a bit at the start, and raising up in bandwidth a bit slow. maybe it's not using cubic congestion control?

when i first tried vps's i found that openvz network performance was often worse than xen for long distances, partially due to them generally using old linux 2.6.18. since then there have been quite a few improvements that help "nz internet".

generally speaking, vps's often seem to do about 3mb/sec, unless nearby/close where it can be 10 to 20 megabytes/sec. a lot of the bigger ones have 10 gigabit pipes through cogent and/or he.net, plus local peering points which should mean "ok" performance from NZ.

Yes, the German box is 100mbit/sec. The reason I am happy to post that one, is that it's not doing anything else with that bandwidth :-)

I'm getting a little stuck for nice machines to test from, so if you have anything, ...

yeah, i have heaps :) can pm you addresses if you want, they all have gig caps so don't really like having them pasted on the net too much.

i can do iperf testing too.

mercutio

1392 posts

Uber Geek

#689418 21-Sep-2012 13:51

quakeguy:
root@de:/var/www# cat /etc/issue
Ubuntu 11.10 \n \l
root@de:/var/www# uname -r
3.0.0-14-generic-pae
root@de:/var/www# cat /proc/sys/net/ipv4/tcp_congestion_control
cubic
root@de:/var/www# cat /proc/sys/net/ipv4/tcp_ecn
2

I'm getting a little stuck for nice machines to test from, so if you have anything, ...

Check /proc/sys/net/ipv4/tcp_wmem.

Linux 3.2 adds proportional rate reduction which reduces the speed slower when there's packet loss (over a whole window instead of instant). It's in the next Ubuntu release from yours. (the LTS)

https://lwn.net/Articles/458610/ (article about the rate reduction)

Linux 3.5 adds fast/early retransmit.

SneakerPimps

104 posts

Master Geek

#689445 21-Sep-2012 14:14

http://149.20.4.69/pub/linux/kernel/v3.0/linux-3.5.4.tar.bz2

1Thread = ~150KB/s
4 Threads = ~550KB/s

http://de2.tw.net.nz/testfile
1 Thread = ~220KB/s
4 Threads = ~820KB/s

The speed fluctuates quite a bit, but it was about the average.