NO02 Poster 105

from ocean to cloud
LOW LATENCY CABLES

Alasdair Wilkie (Hibernia Networks)
email: alasdair.wilkie@hibernianetworks.com
Hibernia Networks
International Exchange Centre
Clonshaugh Industrial Estate
Dublin 17
Ireland
Abstract: Low Latency cables are a new and interesting addition to the submarine cable
community where they are used to support the lowest latency routes between city pairs i.e.
New York and London. Whilst the cables are limited in their routing, there can be only one
"Lowest Latency" path on any one route. In particular they present a series of challenges that
go against the grain with regards to the standard way of thinking with respect to route
engineering and improvements need to be made in laying techniques to overcome the dangers
of untried routes. This paper will provide an insight to the rationale for low latency cable and
explain how Hibernia Atlantic have overcome the challenges and embarked on the lowest
latency path from London to New York including the new Express cable.
1. Why Low Latency?

The main requirement for Low Latency
routes has come from High-Frequency
Traders, sometimes known as Algorithmic
Traders.
Although there are other customers who
are looking for lower latency i.e.: cloud computing
video on demand systems
replication systems
film editing
2. Euclidean, Haversine and Great

Circle Distances.
The principle behind a low latency cable is
the shortest route, and as every school boy
or girl knows the shortest distance between
two points is a straight line. So looking at
London to New York, a straight line using
Euclidean geometry between two points in
space is given by:-
Copyright SubOptic2013
Where :d = distance
R = distance from origin
= Latitude in radians
= Longitude in radians
This gives a distance of 5401 km but has
the disadvantage of tunnelling through the
earth to a depth of greater than 600 km.
This therefore is the first impractical
solution that we come across.
The next solution is to use the Haversine
formula to calculate the Great Circle route
between two points on a sphere, this is
given by :-
Page 1 of 6
from ocean to cloud

3. Latency Calculations
Where :-
This gives us a distance of 5577km from

London to NewYork but the Haversine
formula is for a sphere and the Earth is an
oblate spheroid. Although up until fairly
recently this would have been an
acceptable method, however in 1975
Vincenty(1) derived a formula since
modified by Karney(2) based on work by
Bessel which has significantly improved
the accuracy of Great Circle distances.
The Vincenty formula uses an iterative
calculation until the difference between the
is < 10-12 or 0.6mm accuracy. The full
formula can be found in Ref 1.
Using the modified Vincenty formula the
Great Circle distance from London to New
York is 5594km this therefore is the
theoretical shortest distance that could be
achieved for a cable from London to New
York. This still doesnt take into
consideration any hills or valleys that will
change the length of the route.
A simplistic calculation for latency and

one that works for short routes particularly
on land is that there is 1ms of Round Trip
Delay (RTD) for each 100km of route.
However when applied to transoceanic
cable routes this simplistic calculation is
inaccurate and a more empirical formula is
required.
Breaking down the route into constituent
components we see that the following are
the major contributors to latency
Fibre
Amplifiers
Equalisation
Transponders
Dealing with each of these in turn we have

Fibre
Light travels in a vacuum at 299,792.458
km/s, however in a fibre the glass slows
the transmission of light down and this is
reflected in the refractive index of the glass
which for standard fibre types commonly
used in long haul telecommunications is in
the region of 1.463 to 1.5.
Using a range of these index figures on a
hypothetical 1000km link gives the
following latencies for each:IOR
1.463
1.4682
1.5
Latency for 1000km (ms)

4.88
4.90
5.00
%
Ref
+0.36
+2.53
So even for a small decrease in the

refractive index a useful improvement in
latency can be achieved.
Great Circle route from London to New
York.
For our calculation the unit latency per km

is calculated as follows:-
Page 2 of 6
from ocean to cloud

importantly the latency is the same for all
channels.
Where c is the speed of light in a vacuum
in km per ms.
Amplifiers
The majority of amplifiers on the market
today have internal fibre lengths varying
from 30 to 60m meters in each amplifier.
About 15m of this is for the erbium doped
fibre and the rest is for splicing and
storage. The more amplifiers the more
fibre and the greater latency, so reducing
amplifier count downwards and reducing
the fibre in the amplifiers will again have a
positive effect on the latency.
Amplifier latency is designated as RTDAMP
Transponders
Transponders at the end of each link or
segment provide the final interface to the
customers network. The majority of
interfaces are still at 10 Gbit/s even with
the advent of 40Gbit/s and 100Gbit/s line
rates. Transponder latency varies from
50s to 200s per card. Transponder
latency per transponder pair is designated
by RTDTX
So to add all of the above together we have
an empirical formula for a simple cable
landing station to cable landing station
system as follows:-
Equalisation
There are two areas of equalisation that
need to be considered, one is the inline
equalisation on the submerged plant and
the other is the equalisation of the signals
in the terminal station.
The inline equalisers are a similar prospect
to the amplifiers and since their numbers
are usually very small these can be ignored
for all practical purposes.
Equalisation in the terminal station, is done
in one of three ways, fibre equalisation,
Bragg gratings or for coherent systems
electronic equalisation.
In-line fibre equalisation should be avoided
in low latency systems as not only will it
add significant latency but since it is
carried out on a per channel basis no two
channels will have the same latency.
Where:l is the total fibre length

x is the number of amplifiers
In the empirical formula above we have
ignored any equalisation units either land
or marine based, however this will provide
a good approximation of the achievable
latency.
4. Route Considerations
With all cable routes, and in particular
ones that contain constraints such as low
latency, the consideration for a combined
land and sea route is paramount. This also
means that the integration for the land and
sea routes must be taken into account,
every interface can add latency.
Bragg gratings or electronic equalisation

can both be used as these add insignificant
amounts of latency to the system and
Page 3 of 6
from ocean to cloud

large towns and cities. The great circle
distance from London to Llan-non is
303km whereas the shortest usable road
route is 344km.
The route above from London to New

York would pose some serious challenges
with regards to building this route, the
number of land to sea interfaces as well as
the limitations regarding the land routes
would make this an economically
unjustifiable route to construct.
If the route is split between land and
marine, in the knowledge that the majority
of marine routes can follow a great circle
route if required, or at least close to one,
then we can understand how far from the
great circle route we can deviate before we
compromise the overall latency of the
system. All fibre routes below are based on
using the walking distance between two
points on maps.google.com3, which whilst
not perfect is an easy way to start.
The same issue is also true of looking at

the route across Ireland. The landing point
on the above route is close to Glendoyne
with an exit point near Pollnaclogha. The
great circle route between these points is
221km but the shortest path for a cable
would be 255km.
Doing the same for the other land sections

we have the following (all distances in
km):-
Great
Circle
Route
Shortest
Duct Route
UK
303
344
Ireland
211
255
Newfoundland,
Canada
414
525
PEI & Nova

Scotia, Canada
156
196
USA
356
400
Country
Looking specifically at the section of the

great circle route above, then there are two
particular issues that make this route
economically impossible. One is the direct
route from London to just north of Llannon on the Welsh coast, this direct route
passes through the Brecon Beacon
mountains as well as the centre of many
If these increases in distance are added up

Page 4 of 6
from ocean to cloud

we get an increase in length from the great
circle route of 280km this is for duct, to
realistically look at fibre length then 10%
must be added to the duct length to get the
fibre length, making the variation 308km.
Taking this as a staring point the redesign
using a primarily marine route can take
place.
5. Route Redesign
Since submarine cables can invariably be
installed in almost a great circle route then
increasing the amount of submarine cable
in the route between London and New
York is the most effective way of keeping
the latency low.
For the North American continental shelf

after passing the southern Avalon
Peninsula in Newfoundland the next way
point is the southern tip of Nantucket
Island, before heading to the beach near
Shirley, NY.
Doing this between Weston-super-Mare in

the UK and Shirley in Long Island
provides a great circle route of 5323km so
this is the next starting point for low
latency reduction.
The submarine route shall be split into
three main areas, the deep water section
and the European and North American
Continental shelves.
For the European continental shelf, the
obvious starting point is somewhere at
Weston-super-Mare, UK. The next limiting
point is Fastnet rock on the south west of
Ireland. From Fastnet a great circle route
can be drawn straight to Newfoundland.
Putting all of these waypoints into a trans

Atlantic route we have a route of 5364km
which is only 41km longer than the great
circle route between the two points.
Putting these waypoints into a map format

we can see that he beginnings of a route
would look something like this.
Page 5 of 6
from ocean to cloud

This is the starting point for the route
before the detailed engineering can take
place.
The detailed engineering is the same as
would be done on an ordinary route so the
following must be take into account:
Windfarms and other offshore

renewable energy
Oil and Gas Exploration blocks
Severe slopes
Marine Conservation Zones and
other protected areas
Other cables and pipelines
Limitations
imposed
by
governments within 12 mile limit
and the EEZ
Traffic separation zones
Anchoring areas
Fishing areas
This list is not exhaustive and there will be

other limitations on the route which come
up during the desktop study or survey.
Putting all of these facts into consideration

and
ensuring
that
the
ICPC
recommendations for pipeline and cable
crossings are taken into account a survey
route has been found.
This route together with the existing route

from Halifax to New York and a new route
from Weston-Super-Mare to London will
provide Hibernia Networks with the
latency required.
7. REFERENCES
[1] Vincenty, T. (April 1975a). "Direct and
6. Final Route for Hibernia Express

Due to the fact that Hibernia Networks
already have a cable from Halifax to New
York via Lynn, which is fairly close to the
great circle route from London to New
York, it was decided to bring the cable into
the current landing station near Halifax.
This also had the advantage of avoiding the
very shallow Georges Bank and making
the regulatory burden less by not bringing
a cable into the USA.
Inverse Solutions of Geodesics on the Ellipsoid

with application of nested equations". Survey
Review XXIII (176): 8893.
[2] Karney, C. F. F. (2013). "Algorithms for

geodesics". J. Geodesy 87 (1): 4355
[3] Maps.google.com Google
The other areas which have been avoided

are Ballards Bank to the south of the
Avalon peninsula due to steep slopes. Irish
territorial waters for licensing reasons,
Welsh territorial waters for special areas of
conservation and the Atlantic Array
windfarm area in the Bristol Channel.
Page 6 of 6

NO02 Poster 105

Uploaded by

Copyright:

Available Formats

NO02 Poster 105

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

NO02 Poster 105

Uploaded by

Copyright:

Available Formats

from ocean to cloud

LOW LATENCY CABLES

1. Why Low Latency?

2. Euclidean, Haversine and Great

from ocean to cloud

This gives us a distance of 5577km from

A simplistic calculation for latency and

Dealing with each of these in turn we have

Latency for 1000km (ms)

So even for a small decrease in the

For our calculation the unit latency per km

from ocean to cloud

Where:l is the total fibre length

Bragg gratings or electronic equalisation

from ocean to cloud

The route above from London to New

The same issue is also true of looking at

Doing the same for the other land sections

PEI & Nova

Looking specifically at the section of the

If these increases in distance are added up

from ocean to cloud

For the North American continental shelf

Doing this between Weston-super-Mare in

Putting all of these waypoints into a trans

Putting these waypoints into a map format

from ocean to cloud

Windfarms and other offshore

This list is not exhaustive and there will be

Putting all of these facts into consideration

This route together with the existing route

6. Final Route for Hibernia Express

Inverse Solutions of Geodesics on the Ellipsoid

[2] Karney, C. F. F. (2013). "Algorithms for

The other areas which have been avoided

You might also like