Calculating exact probability distributions for nu, nm and Rs and RD is computationally expensive' W

1
Estimating the Capacity Demand of an IP
Multicast-enabled Network for Transport of TV
Channels
Zlatka Avramova1, Danny De Vleeschauwer1,2,
Sabine Wittevrongel1, Herwig Bruneel1 1 SMACS
Research Group, TELIN, Ghent University,
Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium
kayzlat, sw, hb_at_telin.ugent.be 2 Network
Strategy Group, Alcatel Bell, Copernicuslaan 50,
B-2018 Antwerp, Belgium danny.de_vleeschauwer_at_alc
atel.be
Problem and Assumptions
Approach

• Dimensioning networks for IP-transported TV
  channels distribution estimating capacity
  demand R
• A bouquet of K TV channels is delivered to the
  client in either unicast or multicast mode
• Channel popularity distribution follows ZIPF LAW
• (with parameter a)
• N users (clients, subscribers, customers)
• Probability of a user being active is a
• Parameter ß bandwidth multicast / bandwidth
  unicast channel
• State vector C describing the state of the system
  (its elements ci , where i1 to K, show how many
  users are tuned into every channel)

• The M most popular channels are subject to
  multicast all the time and the rest are subject
  to unicast on request.
• Capacity demand Rs is calculated taking into
  account
• The contribution of multicast channels all the
  time nmM
• The contribution from expected number of unicast
  channels nu.

STATIC CASE
Generalized Network Infrastructure

• All channels can be subject to multicast as well
  as to unicast.
• Capacity demand RD is calculated taking into
  account
• The estimated nm number of multicast channels
• The contribution from expected number of unicast
  channels nu.

DYNAMIC CASE
Methodology
Resource Demand
Calculating exact probability distributions
for nu, nm and Rs and RD is computationally
expensive. We assume the variables have Gaussian
distribution and we verified that the Gaussian
approximation matches the exact results very well
in both dimensioning cases
STATIC CASE
STATIC CASE
DYNAMIC CASE
a mobile TV infrastructure /hybrid network/

IPTV Cable distribution plant TWO EXAMPLES
OF NETWORKS DELIVERING IP-transported TV CHANNELS
DYNAMIC CASE
Results
Optimality of the Static Case
Superiority of the Dynamic Case
Gain is achieved only if Rs lt ßK. For a given
blocking probability Pblock there is an optimal
M which minimizes the capacity demand
Rsf(M). Depending on the settings
there may or may be not capacity gain. The
function Rsf(M) when varying the parameter a of
the Zipf distribution or the number of users N is
shown
The dynamic dimensioning scheme always
outperforms the static dimensioning approach.
Capacity demand R increases logically with
increasing number of users N. In both scenarios
we consider that as long as Rs or RD is less than
ßK (300 in the example below), it is feasible to
employ them and thus achieve capacity gain
otherwise, when Rs or RD ßK, there is no
benefit in deploying such dimensioning schemes
but rather multicast (broadcast) all channels.
As long as number of users N is
smaller than approximately 600, there is a
certain gain in both cases. If number of users is
more than 600 but still less than approximately
1350, only the dynamic dimensioning scheme leads
to capacity saving and if N is more than 1350,
none of both schemes is justified to be deployed.
The limit ratio N/K of applicability region
of both scenarios is shown also for ß1 2 and
for two values of a1 2. It proves again the
superiority of the dynamic case over the static
case.
If N is 200 and M is chosen between 20 and 30, a
capacity gain of 44 is achieved. If N changes
drastically to e.g. 600, Rs in a mixed
unicast-multicast static case surpasses 150
channels (corresponding to the case when all
channels are broadcast (multicast)). Otherwise,
if M is chosen larger than 60 and N becomes 600,
there is still some capacity gain.
If a a3 and if M is chosen between 50 and 70,
Rs is 150 in a mixed unicast-multicast static
case, while in an all-multicast case, R is 200 gt
a capacity gain of approximately 25 is achieved.
Conclusions
Digital television distribution over IP networks
can benefit from multicasting the most popular
channels. We consider two cases a static case in
which the number of channels to multicast is
predefined and a dynamic case where all channels
can be subject to unicast or multicast depending
on the momentary behavior of the viewers. We
showed that the total capacity required, i.e.,
the contribution of the unicast and of the
multicast demand, is approximately described by a
Gaussian variable in both cases. We use this
Gaussian approximation to determine the optimal
threshold which channels to multicast in the
static case. We further study the impact of the
different parameters (number of users, activity
grade, number of channels and shape of popularity
distribution of channels) on the resource demand
function and the optimal multicast threshold. We
compare the static case with optimal settings to
the dynamic case. Specifically, we determine for
which parameters the cases still result in a
capacity gain. We prove that the dynamic case
always outperforms the static one. Network
operators can use the methodology presented to
estimate the capacity demand in multicast-enabled
networks and the gain achieved over all-unicast,
all-multicast or hybrid networks.
References
kayzlat_at_telin.ugent.be http//telin.ugent.be/smacs
TELIN Department SMACS Research Group