'TD-Sense Algorithm'-A Novel Approach to Improve Network Performance for 1G-EPON

Abstract

This paper proposes a novel ‘TD-sense algorithm’ for
improve the network performance and compares it with a
standard DBA_GATED algorithm. The algorithm is implemented
by emulating an access network with the use of two 1G
transceivers for 1G EPON. The algorithm is tested for Triple Play
services that include simulated voice, video and data packets. The
algorithm is found to maintain a better tradeoff between
throughput and delay, than the existing DBA_GATED algorithm.

Full text

Journal of Network Communications and Emerging Technologies (JNCET) www.jncet.org
Volume 3, Issue 1, July (2015)
ISSN: 2395-5317 ©EverScience Publications 64
‘TD-Sense Algorithm’-A Novel Approach to
Improve Network Performance for 1G-EPON
Satya Prakash Rout
Department of Physics, Sri Sathya Sai Institute of Higher Learning, Puttaparthi, Andhra Pradesh, India
Abstract – This paper proposes a novel ‘TD-sense algorithm’ for
improve the network performance and compares it with a
standard DBA_GATED algorithm. The algorithm is implemented
by emulating an access network with the use of two 1G
transceivers for 1G EPON. The algorithm is tested for Triple Play
services that include simulated voice, video and data packets. The
algorithm is found to maintain a better tradeoff between
throughput and delay, than the existing DBA_GATED algorithm.
Index Terms – OLT, ONU, Triple Play, Average Delay,
Throughput, REGISTER, QOS, MTU
1. INTRODUCTION
Passive optical networks (PON) were developed to satisfy the
demands of high bandwidth and larger coverage needs of future
access networks [1]. A Passive Optical Network is a point-to-
multipoint network that has no active elements in the path of
the signals from source to destination [2]. The emergence of
PON has reduced the cost to transmit data from Optical Line
Terminal (OLT) which is located at the central office (CO) to
Optical Network Unit (ONU) that provides broadband internet
access to the subscribers. Ethernet PON (EPON), designated
as IEEE 802.3ah, carries data traffic encapsulated in Ethernet
frames over PON based network. 1G – EPON supports
transmission and reception of data packets at the rate of 1
gigabits/second. The OLT uses Dynamic Bandwidth
Allocation (DBA) to allocate bandwidth dynamically to the
ONUs depending on the current traffic load at the ONU. We
propose a novel algorithm for bandwidth allocation named ‘TD
(Throughput Delay)-Sense algorithm’ wherein the OLT
classifies the data traffic into two different types and allocate
the bandwidth according to priority. This is implemented by
using two 1G transceiver cards fixed to two computers
connected by fiber optic cable, one acting as OLT and the other
containing multiple ONUs. We have compared the
performance of the novel algorithm with an existing algorithm.
2. DBA_GATED ALGORITHM
PON architecture comprises of an OLT located at a Central
Office, connected to various ONUs through Optical Splitter.
OLT acts as an interface between the core and the access
network. The downstream channel from the OLT to the ONUs
employs a single transmitter at the central office
and the upstream channel is shared among the various ONUs
[1]. The OLT broadcasts the traffic in the downstream channel
to all ONUs. DBA algorithm is used as medium access control
mechanism which is employed to achieve the best possible
usage of the bandwidth for the upstream channel usage from
the ONU to the OLT. Each ONU calculates its bandwidth
requirement as per its buffer contents. Then it sends a request
frame to the OLT with the desired transmission window
request. The OLT receives multiple requests from different
ONUs at the same time. It then uses DBA algorithms to
calculate the amount of bandwidth to be allotted to each ONU
and sends the grants to each ONU, after which each ONU
transmits in its own time window.
Multi point control protocol (MPCP) controls the upstream
multiple access to allow efficient transmission of data. Five
control messages namely GATE, REPORT, REGISTER_REQ,
REGISTER and REGISTER_ACK are used for auto discovery and
registration of the ONUs. [3]
In DBA_GATED algorithm, the OLT grants the preferred window
size to each ONU. Further, if the same ONU has to send more
data, it has to send another request to OLT. Though this
algorithm provides a high level of QOS, the drawback is that
an ONU with more data will block the other ONU with less
data. This means ONU with fewer amounts of data has to wait
for a long time until the ONU with more data completes its
transmission. So in DBA_GATED algorithm, a single user is
allowed to meet his demands and standard of service at the
cost of other users, which is inequitable. [4]
3. TD-SENSE ALGORITHM
This is a novel algorithm proposed to gauge the performance
of the network in terms of variation of both throughput and
delay. This is implemented for Triple Play services where delay
sensitive and delay non-sensitive applications exist. Video and
voice over IP constitute the delay sensitive applications and
other multimedia constitute the delay non-sensitive
applications.
Delay sensitive application s typically range from 64 to 2500
bytes and are denoted as Type 0 traffic. Delay non-sensitive
applications like huge video and data files extend from 5000 to
9000 bytes and are termed as Type 1 traffic. In this algorithm,
the report message of all users is stored inside the buffers of the
ONUs. The report message may be delay sensitive or non-
sensitive.

Journal of Network Communications and Emerging Technologies (JNCET) www.jncet.org
Volume 3, Issue 1, July (2015)
ISSN: 2395-5317 ©EverScience Publications 65
Let T1 = time when ONU sends a report to the OLT.
T2 = time when the report is received at the OLT.
T (0) = Report length of the delay sensitive traffic in bytes or
Type 0 traffic
T (1) = Report length of the non-delay sensitive traffic in bytes
or Type 1 traffic
Channel free time = T2 - T1
Four possibilities exist in this approach. The following cases
explain how scheduling will be done based on priority.
Case 1: When T (0) > T2 – T1, then T (0) will be scheduled and
T (1) will not be scheduled. This condition indicates that delay
sensitive traffic should be scheduled immediately and keeps the
non-delay sensitive in buffer to be scheduled in the next cycle.
Case 2: When T (0) < T2 – T1, then T (0) will be scheduled and
T (1) will be scheduled at the channel free time.
Case 3: When T (0) + T (1) < T2 – T1, then both T (0) and T (1)
will be scheduled.
Case 4: When T (0) + T (1) > T2 – T1 and T (0) < T2– T1, then
both T (0) and T (1) will be scheduled but report length of T
(1) will be cut down so that it can be accommodated in the
channel free time.
The traffic generator module generates Type 0 traffic at the rate
of ß1 in Gbps and Type 1 traffic rate at the rate of ß2 in Gbps.
The offer load ß is given as
ß = b1 ß1+ b2ß2 (1)
Where b1 = the mean packet size of Type 0 traffic varying from
64 to 1500 bytes.
b2 = the mean packet size of Type 1 traffic varying from 5000-
9000 bytes.
4. NETWOK DEIGN AND TOOLS
The set-up used two 1G Ethernet server adapters which are
capable of handling data speeds up to 1.25 Gbps giving a real
time functionality. The network uses optical fiber patch cords.
In order to emulate a passive optical network two Linux loaded
machines, one as OLT (Server Side) and the other as having
multiple ONUs in it. Since this resembles a point-to-point
network connection rather than a PON, the ONUs and the end
nodes are to be simulated so as to follow the PON architecture
as shown in the figure [1].
Figure 1: Network Simulation Design
The simulation was carried out by creating multiple processes
which act as ONUs and each process has multiple threads
running that act as end nodes. The multiple threads randomly
generate the traffic. This is similar to the end nodes that
generate real time traffic. Thus the entire setup of a passive
splitter and the ONUs at different distances is simulated on a
single system and the OLT on the other system.
A traffic generator module also known as random packet
generator (RG) is used to generate packets and pump them into
the network. The rate at which packets are generated is kept
random so as to achieve the closest approximation to the real
time scenarios. Two traffic generator modules are used in the
simulation. The mathematical equation followed for
generation of random packets is given as
X = - log (1- µ) ÷ β1 (2)
Where µ is the random number generated by the system and
β1 is the rate parameter varying from 0 to 1. This traffic
generator module follows the exponential function.
The second traffic generation module implemented is based on
Pareto Distribution that generates ON - OFF periods. The
Pareto distribution is a heavy tailed distribution with a
probability density function given by [6]
F(x) = β2 * bβ2 ÷ Xβ2 +1 (3)
Where β2 is the shape parameter and b is the location
parameter. The number of packets per second (ON period)
follows the Pareto Distribution with the rate factor varying
from 0.2 to 0.8. The OFF period also follows Pareto
Distribution.
Using the above mathematical equations the packet size can
be varied using the rate factors β1 and β2. Similarly packets of
random lengths are generated in the range of 64 and 9000
bytes. Therefore each simulated thread, generates packets of
Threads Processes Threads Processes
RG 1
111
RG 2
22
RG 1
RG 2
LINUX INSTALLED PC 1
ONU
ONU
OLT
LINUX INSTALLED
PC 2
OPTICAL FIBER

Journal of Network Communications and Emerging Technologies (JNCET) www.jncet.org
Volume 3, Issue 1, July (2015)
ISSN: 2395-5317 ©EverScience Publications 66
random length at random intervals of time. At the ONU, all
the packets were collected and encapsulated into a packet size
of 9000 bytes and sent to the OLT. We have used the MTU
(Minimum Transfer Unit) to be 9000 bytes. This aids the
transfer of packets at higher rates.
5. RESULTS AND DISCUSSION
The results of DBA_GATED and TD-sense algorithms were
measured by considering various simulation parameters like
delay, data throughput, offered load, traffic type and number of
ONUs connected to OLT. Eight ONUs are simulated in one of
the systems while the other system has OLT running in it.
Exchanging of control messages between OLT and ONU plays
an important role for synchronization. It is found that, when the
transmission process is on, the low priority control messages
get dropped at the socket level of OLT. This occurs because,
the OLT, when ready to send data to an ONU1, receives a
control message from ONU2. Since the transmission is in
progress, the OLT drops this control message from its buffer.
As a result, ONU2 starts waiting for the reply and at the same
time OLT also waits for the control message from ONU2 which
it had sent already. This gives rise to confusion, as OLT and
ONU are waiting for each other’s responses. To tackle this
problem, we modified the communication between OLT and
ONU at their socket level. OLT was made to use two sockets
of different port numbers to communicate to a particular ONU
wherein one socket deals with control message and the other
port deals with data transfer. ONU uses the same socket for
control message and data.
Packets are generated by various ONUs and transmitted to the
OLT. A network bandwidth testing software, ‘Netperf’ was
used to determine average throughput, bytes transmitted and
the bytes received with a corresponding delay when the packets
move from various ONUs to the OLT. The transceiver cards
use 1490 nm as upstream wavelength and 1310 nm as
downstream wavelength. The graphs given here (Fig. 1 & 2)
show how the TD-sense algorithm is better in performance than
the DBA_GATED algorithm. In DBA_GATED algorithm,
OLT allocates the time window for the ONU’s demands
without considering whether the data is delay sensitive or not.
But in case of TD-Sense algorithm, the ONUs are allocated
timing windows by OLT as per the report lengths stored in
buffer. The data traffic present in the buffer are classified into
delay sensitive (Type 0) and delay non sensitive (Type 1). If in
a cycle, the buffer contains more number of report lengths of
Type 0 traffic than Type 1, then Type 0 utilizes the channel free
time. This would lead to Type 1 traffic waiting for subsequent
cycles for accessing the channel free time, thus creating
bandwidth starvation. TD-sense algorithm solves this by
imposing the condition that if Type 0 traffic repeats many times
for a given number of cycles, then in the next cycle, the TD-
sense algorithm makes Type 1 traffic use the channel free
time. Thus it is well suited for both delay sensitive applications
like Voice over IP, Video on Demand and delay non-sensitive
applications like huge data transfers between users.
Figure 2: Throughput (Gbps) vs Offred Load (Gbps)
Figure 3: Average Delay (ms) vs Load (Gbps)
Given an offer load, TD-sense algorithm gives better
throughput than DBA_GATED algorithm. It can be seen from
Fig.1 that when delay sensitive traffic is more i.e., when Type
0 is 80% and Type 1 is 20%, the throughput is maximum in this
case. There is a decrease in the throughput when the delay non-
sensitive traffic is more i.e., Type 1 is 80% and Type 0 is 20%.
In both the cases, TD-sense outperforms the DBA_GATED
algorithm.
In DBA-GATED Algorithm the delay increases with increase
in offer load. This happens because whatever window size an
ONU requests, OLT grants it fully. In case of TD-Sense

Journal of Network Communications and Emerging Technologies (JNCET) www.jncet.org
Volume 3, Issue 1, July (2015)
ISSN: 2395-5317 ©EverScience Publications 67
Algorithm, the delay is the average delay which is the sum of
ONU processing delay, OLT processing delay and Round Trip
Delay. It can be seen in Fig.2, that the TD-sense algorithm
minimizes the delay when the load is delay sensitive (i.e., Type
0 is 80%) and delay increases when the traffic is not delay
sensitive (Type 1 is 80%).
Thus TD-sense algorithm maintains a tradeoff between
throughput and delay by classifying the data traffic as delay
sensitive and delay non-sensitive. However, we have found
that, when the offer load is more than 1 Gbps, the throughput
seems to be decreasing, the cause of which is to be explored.
This may put a limitation on its applicability for higher offer
loads.
6. CONCLUSION
This paper presents a novel ‘TD-Sense algorithm’ for Triple
Play Services by simulating semi real access networks with the
use of two 1G transceivers and simulating OLT and ONUs in
two different systems. It is found that this algorithm maintains
a tradeoff between throughput and delay. Bandwidth starvation
is overcome in this algorithm. Also, it is more efficient than the
DBA_GATED algorithm in terms of the throughput and delay.
Real time video relaying through the network and measuring
the network efficiency with the increasing number of ONUs
can be further explored.
REFERENCES
[1] Neeharika Jana, Ganesh C. Sankaran, Krishna M.Sivalingam and Gerard
Parr, “Performance analysis of dynamic bandwidth allocation algorithms
for long reach PONs”, IEEE 4th International Symposium on Advanced
Networks and Telecommunication Systems,Mumbai,IEEE,3 pp., 2010
[2] A. S. Reaz, L. Shi, and B. Mukherjee, “Broadband access networks: Cur-
rent and future directions,” in Next-Generation Interne
t:
Architectures
and Protocols, B. Ramamurthy, G. Rouskas, and K. M. Sivalingam,
Eds. Cambridge University Press, 2010.
[3] A. Mady and A. Tonini, “A vhdl implementation of onu auto-discovery
process for epon,” in Networking and Media Convergence, March 2009.
[4] GW Technologies Co., Ltd., www.gwtt.com, EPON Technology White
Paper.
[5] Gutierrez and P. Garfias, “Next generation optical access networks:
from tdm to wdm,”Trends in Telecommunications Technologie, 2010
under CC BY-NC-SA 3.0 license.
Author
Mr. Satya Prakash Rout received the bachelor
degree in Electronics & Communication
engineering from Gandhi Engineering College,
Bhubaneswar, in 2010. He is also received his
master degree in Fiber Optics and Digital Image
Processing from Sri Sathya Sai Institute of Higher
Learning, Puttaparthi, Andhra Pradesh in 2013.
Currently, He is an Assistant Professor at
TempleCity Institute of Technology and
Engineering, Khurdha. His interests are Digital
Communication, Computer Networks, Wireless Computing and Optical
Communication system.
.