Learning about the internet

Mailman List Mirror (Read Only)
1

Hi,

I was just reading about transatlantic cabling in some hopes that I
would be able to find an answer as to why the latency between here in
greece and Los Angeles is roughly ~250ms. This seems to be a really
common thing, although I'd like to understand why and the articles on
transatlantic cabling as near as I can tell indicate that I am getting
screwed if anything (not enough information?)

(from Los Angeles to my house)
Konsole output

Konsole output
gw~ #mtr --report-wide xxxxxxxxxxxxxxx.access.hol.gr
Start: Mon Nov 3 13:04:02 2014
HOST: gw Loss% Snt Last Avg
Best Wrst StDev
1.|-- 208.79.92.65 10.0% 10 1.5 3.6
  1.2 15.5 4.6
2.|-- s7.lax.arpnetworks.com 0.0% 10 0.8 10.9
  0.8 54.2 20.7
3.|-- vlan953.car2.LosAngeles1.Level3.net 30.0% 10 10.5 10.3
10.1 10.8 0.0
4.|-- ae-27-27.edge6.LosAngeles1.Level3.net 30.0% 10 21.8 16.2
  8.6 47.6 14.7
5.|-- ae-4-90.edge1.LosAngeles6.Level3.net 80.0% 10 9.0 8.9
  8.7 9.0 0.0
6.|-- be3036.ccr21.lax04.atlas.cogentco.com 10.0% 10 1.7 2.1
  1.4 4.3 0.7
7.|-- be2076.mpd22.lax01.atlas.cogentco.com 10.0% 10 1.6 1.9
  1.6 3.2 0.0
8.|-- be2068.ccr22.iah01.atlas.cogentco.com 0.0% 10 37.7 37.7
37.3 39.0 0.3
9.|-- be2173.ccr42.atl01.atlas.cogentco.com 0.0% 10 51.6 52.4
51.5 57.5 1.7
10.|-- be2171.mpd22.dca01.atlas.cogentco.com 0.0% 10 62.6 62.7
62.4 63.3 0.0
11.|-- be2112.ccr41.iad02.atlas.cogentco.com 0.0% 10 155.5 155.8
155.5 156.1 0.0
12.|-- be2268.ccr42.par01.atlas.cogentco.com 0.0% 10 152.6 152.7
152.5 153.5 0.0
13.|-- be2278.ccr42.fra03.atlas.cogentco.com 0.0% 10 155.3 155.4
155.1 155.5 0.0
14.|-- be2229.ccr22.muc01.atlas.cogentco.com 0.0% 10 161.2 161.1
160.9 161.3 0.0
15.|-- be2223.ccr21.vie01.atlas.cogentco.com 0.0% 10 164.9 165.1
164.9 165.2 0.0
16.|-- be2046.ccr21.sof02.atlas.cogentco.com 0.0% 10 189.5 189.4
189.3 189.9 0.0
17.|-- be2118.rcr11.ath01.atlas.cogentco.com 0.0% 10 197.5 197.6
197.4 197.7 0.0
18.|-- 149.11.120.38 0.0% 10 202.7 202.2
200.3 204.2 1.4
19.|-- 62.38.97.113 80.0% 10 208.5 209.8
208.5 211.1 1.7
20.|-- gigaeth04-13.krs00.ar.hol.gr 60.0% 10 211.3 213.0
211.2 218.2 3.4
21.|-- ??? 100.0 10 0.0 0.0
  0.0 0.0 0.0
22.|-- xxxxxxxxxxxxxxxx.access.hol.gr 40.0% 10 231.3 231.4
231.2 231.7 0.0
gw~ #

And to be more clear: I am hoping to learn about the complex trials that
these packets are going through and how time is being lost if the
latency across the transatlantic cable is really capable of less the
60ms of latency? Sure over capacity (3.2Tbits/s wow jeez) is one answer,
but what are some other possibilities for loss of time?

Also it seems with my VPN (OpenVPN) tunnel I get the most reliable
connection (fewest drops) with:

Konsole output
mssfix 576
fragment 576

Although this could be a false positive as it only *seems* to help with
reliability since I changed it. Even then but less often than before I
still experience drops but I want to believe that's possibly due to my
ISP at that point.. but assuming my ISP was absolutely perfect and never
a problem what else there to consider?

Any and all insight is appreciated.

-Paige

2

looks about right in the neighborhood of 9k miles...

from lax or therebouts.

  Upstream Intf Nexthop Sent Loss Min Avg
     Max Dev
  cogent x x 10 0.0% 194.814 210.255 240.989
16.518
  comcast x x 10 0.0% 201.723 213.942 239.47
13.205
  l3 x x 10 0.0% 195.51 208.189 226.971 10.455
  telia x x 10 0.0% 194.552 207.392 225.792 10.321

3

^^ There's the largest part part of your answer. IAD and DCA are both
Northern VA airports; there should be perhaps 4ms of latency between these
locations.

A more smug answer is, "Cogent." Google "cogent congestion," "Cogent
peering," and similar search terms.

Other parts of the answer include:

1. The routers are mostly doing store and forward which means each must
receive your entire packet before it can forward it onward to the next
router. This will take more or less time depending on the link speed.

2. There are layer 2 devices doing the same thing in between some of the
routers. Layer 2 is invisible to the layer 3 traceroute. There may also be
MPLS systems in there with the same impact.

3. Greece and LA are more than 11,000 km apart as the crow flies. This
means round-trip packets travel something like 25,000 km. There is some
signal propagation delay involved. U.S. East Coast to Hawaii (about 80% of
that distance) is typically 120 ms.

Regards,
Bill Herrin

4

Hey Paige,

That¹s going to be extremely common when traversing transatlantic cable.
Hopefully I can help explain.

Most of your latency comes from simple speed of light limitations through
optical fiber (~35% slower than vacuum) combined with the fact that the
fiber must take a somewhat indirect path to your destination.
Additionally, a small percentage of latency will be added for all of the
switching/routing/regeneration nodes that must convert the fiber signals
from optical back to electrical (for processing / regeneration) then back
to optical. In your case, your packets are taking the following path:

Los Angeles (Arp Networks to Level3 to Cogent) -> Houston -> Atlanta ->
Washington DC -> Paris -> Frankfurt -> Munich -> Austria -> Bulgaria ->
Athens -> Hellas

Essentially it¹s taking 60ms round trip to get to DC, then another 95ms to
get to Paris, plus 60ms to get to Hellas, then another 20ms to traverse
the local ³last mile² access network to your destination.

If you had a *direct* fiber from LAX to Hellas Greece, that would be
~22,000km of round trip fiber. Working out some simple math, that would
yield approximately 110ms of round trip latency no matter how much
bandwidth you have. Now, add in all of the indirect fiber paths through
major cities plus all of the routing/switching hops and you can see why
you¹re getting 200ms+ latency.

Hope this helps!

Thanks,
Dustin Melancon, Sr. Network Engineer
225-214-3864 Direct ::: 866-978-3698 Toll-Free
­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­
VENYU : Your Data Made Invincible : www.venyu.com <http://www.venyu.com/>
­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­­
Read our latest blog post: blog.venyu.com <http://blog.venyu.com/>

5

Isn't this most likely a side effect of MPLS tunneling and the 93ms jump
there is actually the trans-atlantic segment? All the european hops
following it are very close in latency to hop 11.

6

Yeap, that's what I was typing up.

More than likely what we're seeing is just the far of the link from IAD to Paris. This is evident from the fact that the next hop to Frankfurt is only an additional 3ms beyond what is reported as IAD.