IJMTST 2016 | ISSN: 2455-3778 Traffic and Power reduction Routing Algorithm for NOC Cores

98 International Journal for Modern Trends in Science and Technology
Volume: 2 | Issue: 06 | June 2016 | ISSN: 2455-3778IJMTST
Traffic and Power reduction Routing
Algorithm for NOC Cores
Kalivarathan1
| Naveen2
1PG Scholar, Department of Electronics and Communication Engineering, Gojan School of Business
and technology, Chennai, TamilNadu, India.
2Assistant Professor, Department of Electronics and Communication Engineering, Gojan School of
Business and technology, Chennai, TamilNadu, India.
With the progress of VLSI technology, the number of cores on a chip keeps increasing, Now a days we are
increasing the processing level of the chip, NOC is a best method to interconnect the core with each other core
on the chip, it reducing the overall chip power and Traffic level by sharing the work load with other cores on
the chip. And Dynamic Voltage Frequency Scaling (DVFS) is the technique for monitoring the
Frequency/Voltage level of each core of the chip and providing sufficient power to the cores, ATPT is a Table
that having (low and high) Frequency level table of the Each core. ATPT has very high prediction accuracy
system. Depends upon the data speed of the core the voltage/frequency will be given by DVFS. If the core is
in ideal state for a while, that core is moved to low power mode. So the power of the each core will be reduced.
KEYWORDS: Network on Chip, Dynamic Voltage Frequency Scaling, Application Traffic Prediction Table,
Last Value Prediction, Pattern Oriented Predictor, Performance Monitoring Unit, Core Traffic
Copyright © 2016 International Journal for Modern Trends in Science and Technology
All rights reserved.
I. INTRODUCTION
With the progress of VLSI technology, the
number of cores on a chip keeps increasing, now
days we are increasing the processing level of the
chip, NOC is a best method to interconnect the core
with each other core, Here reducing the overall chip
power and Traffic level by sharing the work load
with other cores on the chip. And Dynamic Voltage
Frequency Scaling (DVFS) is the technique for
monitoring the Frequency/Voltage level of each
core on the chip and providing sufficient power to
the cores, ATPT is a Table that having (low and
high) Frequency level table of the Each core. ATPT
has very high prediction accuracy system.
Analyses also show that it incurs a low area
overhead and is very feasible. Increasing the clock
frequency to increase performance is no longer an
option due to, amongst others, energy
consumption, heat developments and the
enormous costs for new technologies. Power and
thermal distribution are two critical problems in
current chip design. With the progress of VLSI
technology, the number of cores inside one chip is
still increasing, so from multi core to many core,
the Network on Chip (NoC) is the most widely used
solution to interconnect the cores. and explicit tile
to tile communication is supported. When cores
are interconnected by the NOC, additional
problems need to be considered, such as power
consumption and heat from switches the control of
traffic and intercommunication timing between
tasks. Apparently, the NoC plays an important role
in the many core design; the increasing number of
cores requires a communication system different
from a conventional bus system, since a bus
quickly becomes the bottleneck of the system. One
approach is to employ a Network on Chip (NoC).
With the ongoing trend to increase the number of
cores on chips, the NoC becomes an essential part
of the system. There are many NoC topologies such
as meshes, trees, multistage interconnection
networks (MINs), and many more. NoCs have
several advantages such as scalability and
modularity.
ABSTRACT

Traffic and Power reduction Routing Algorithm for NOC Cores
Figure 1: Architecture of NOC
According to this observation, in this paper we
propose a novel application driven approach for
predicting traffic in NoC and performing DVFS on
communication links. We consider message
passing many core architectures, in which cores
communicate with each other directly through
explicit message passing. The basic idea is to
capture the communication patterns between
parallel tasks, i.e., the endtoend traffic, by using a
small table in the network interface (NI) of each
core to record the outgoing messages from that
core. The novel data structure is called the
Application driven Traffic Pattern Table (ATPT).
With the support of ATPTs, the amount of data
injected into the NOC from each core can be
predicted. Once the predictions are made, the
utilization of each individual link can also be
derived. The voltage/frequency (VF) level of the link
can thus be adjusted proactively based on the
predicted link utilization.
II. BACKGROUND AND PREVIOUS WORK
A. Networks on Chip
The solution to scalably and efficiently connect
onchip components is a packet switched on chip
network. A NoC consists of a high speed router at
each node, connected by links to its neighbors.
Although the NoC may carry various kinds of traffic
(e.g. interrupt requests), its most important
purpose is to service cache miss requests. At the
heart of this line of research are micro architectural
questions such as router/link design or efficient
topologies. However, as we discuss in this paper,
some of the key problems in on chip networks are
in fact networking problems, rather than
architectural problems.
B. NoC Characteristics
NoCs feature a number of special characteristics:
high performance demands, coupled with
hardware implementation constraints, lead to a
different trade off space for NoCs compared to most
traditional off chip networks. NoCs run at higher
utilization, and traffic patterns do not exhibit flow
character (such as in the Internet), but are
characterized by the self throttling nature of
applications on the various cores. Aspects such as
chip area/space, power consumption, and
implementation complexity (e.g. the expense of
arbitration and routing logic) are first-class
considerations. These and other characteristics
lend on chip networks an interesting and unique
flavor, and have important ramifications on the
resulting networking solutions.
The presented work is completely Depends upon
the traffic prediction, There is no other activity for
reducing the traffic, In NOC cores dynamic voltage
frequency is the major role that can monitoring the
data transmission from one core to another core
and also it can increase/ reduce the power depends
upon the data transmission rate(Bits/Sec). So the
total chip power is reduced. By this system
Application Driven Traffic Prediction Table (ATPT)
is the table that having Low and High frequency
value of each cores. ATPT having two types of table
(i) Last Value Prediction (ii) Pattern Oriented
Predictor, In that the core working in high
frequency that is the core data transmission
speed(Bits/Sec) is high, That Time DVFS provides
sufficient power to the core which is acting high or
low Data transmission rate, DVFS is the best
method to monitor the transmission rate of the
cores, The total power is reduced by operating the
cores as the above method, The overall process is to
monitoring the Traffic level and giving sufficient
power to the cores, There is no other activity to
reduce the traffic level of the system.
C. Limitations
 Predicting the traffic of each cores
 There is no other work to reduce the traffic
level in the chip
III. SYSTEM MODULE
NOC system is the best method for
interconnecting the core with each other cores, The
main contribution is power and traffic, In this
proposed system we are reducing the overall traffic
level and overall power consumption of the chip
.Here the End to End data transmission has
monitored by Dynamic Voltage Frequency Scaling
(DVFS), Depends upon the data transmission rate
the power has given to the cores, This is the
method to manage the wastage power, so the total

chip power as reduced. If the data rate is high that
core considering as a overload system, that time
the data’s are shared with other core on the same
chip, By sharing the data’s with other core, we can
manage the traffic level. So there is no need to give
more power to the core like the existing system.
Figure 2: Block Diagram of Proposing Methodology
We are reducing the Traffic and power level of
the chip, By using DVFS technique , And there are
two types of (high and low) frequency table for
predicting the Data rate of each cores, If the data
rate is high that time the data has been shared
with other core so the traffic level is reduced, To
solve the power consumption problem in NoC,
previous works have proposed techniques such as
dynamic voltage/frequency scaling (DVFS) to
adjust the power mode of the switches and links to
match the traffic flows. The challenge is to predict
the traffic flowing through the switches and links in
the next time interval. For example, if a switch or a
link can be predicted to be idle for a while, then it
can be set to a low power mode by lowering its
voltage or frequency. According to this observation,
in this paper we propose a novel application driven
approach for predicting traffic in NoC and
performing DVFS on communication links.
We consider message passing many core
architectures, in which cores communicate with
each other directly through explicit message
passing. The basic idea is to capture the
communication patterns between parallel tasks,
i.e., the endtoend traffic, by using a small table in
the network interface (NI) of each core to record the
outgoing messages from that core. The novel data
structure is called the Application driven Traffic
Pattern Table (ATPT). With the support of ATPTs,
the amount of data injected into the NOC from each
core can be predicted. Once the predictions are
made, the utilization of each individual link can
also be derived. The voltage/frequency (VF) level of
the link can thus be adjusted proactively based on
the predicted link utilization. In comparison to
previous studies that make the DVFS decision
based on the hardware status, our approach uses
the data transmission behavior of the application
as the VF scaling reference. The data transmission
behavior is a better guide, because it is more
predictable and the repetitive behavior exists in the
execution phase.
Merits
 By sharing the overload data’s with other
core, we can manage the traffic level. So the
overall chip traffic level is reduced
 Depends upon the data rate the power as
given to the cores
 If the core has been idle for a while, then the
core is set to be a low power mode
IV. OVERVIEW OF NETWORK ON CHIP
Today buses are the dominating technology for
systems on chips (SoCs) However, buses have
severe limitations that become evident, if the
number of components in a system is large, The
bus is a communication bottleneck, bandwidth is
limited. Since they provide a much larger amount
of communication resources and are scalable
Network on chip is an emerging paradigm for
communications within large VLSI systems
implemented on a single silicon chip. The layered-
stack approach to the design of the on chip inter-
core communications the networkonchip (NOC)
methodology. "In a NoC system, modules such as
processor cores, memories and specialized IP
blocks exchange data using a network as a "public
transportation" subsystem for the information
traffic. A NoC is constructed from multiple point-
topoint data links interconnected by switches,
such that messages can be relayed from any source
module to any destination module over several
links, by making routing decisions at the switches.
A NoC is similar to a modern telecommunications
network, using digital bit packet switching over
multiplexed links.
Our approach is different from all these works
since, we use dynamic frequency modulation of
routers to manage congestion and improve
throughput. Recently, Moscibroda. show that a
buffer less routing approach leads to lower power
consumption without significantly sacrificing
performance in onchip networks.

Figure 3: NOC Grid Topology
An activity based power management scheme
was recently implemented in the routers of the Intel
80core chip For this chip, power reduction was
achieved by deactivating ports in the onchip router
and putting individual queues in the ports to sleep
or clockgating them based on activity inside the
router. Ours approach is orthogonal to this idea
and can be implemented on top of this design for
performance and power management as well.
A. Link Delay and Power
To assess the impact of global wires, we have
studied 65 nm NoC links in isolation from the NoC
modules. Several factors have to be considered in
link design, includingobviously length and desired
clock frequency. Short or slowclocked links do not
pose problems. However, as either length or target
frequency is increased, an undesired rise of power
consumption is also observed.
These results show that NoC links implemented
using the LPHVT library are substantially more
powereffective, but impose much tighter
constraints on link feasibility.
B. NOC Components
According to the NoC approach, a core is
connected to a switch via a network adapter. The
connection among switches is implemented with
links and determines the logical network topology.
It is worth mentioning that nowadays fabricated
chips are incorporating 8x8 cores (TILERA Tile64).
The structure of a Mesh topology is shown in Fig
.The core is the NoC component which sends and
receives messages. The network adapter relieves
the core from taking care for the communication
process. It is the interface which connects the core
to the NoC. The main task of the adapter is to
handle the flow control, which refers to the
allocation of network buffers and links among the
data competing for network resources in order to
be delivered at their destination. The switch
determines the path over which the packets are
delivered. The basic micro architecture of a switch
is shown in Figure 4.
Figure 4: Architecture of Switch
1. packet routing Link
2. usually bidirectional
3. Resource Network Interface (RNI, NI)
4. Packet sizes data
5. EndtoEnd flow control
6. Processing Element (PE), Core,
7. Resource could be a memory block
The number of inputs of a switch depends on the
topology, considering the topology in the switch
has at maximum four inputs from other switches.
An additional input is needed to enable the receipt
of data from the local core. Incoming data are
stored in local buffers. Note, a switch can also be
output buffered or have both input and output
buffers. According to the routing strategy (Chapter
2.3.2) the router determines the output for
incoming packets. Packets are switched to the
selected output and are forwarded to the adjacent
switch Routing on NoC Routing on NoC is quite
similar to routing on any network. A routing
algorithm determines how the data is routed from
sender to receiver. Routing algorithms are divided
into two groups, oblivious and adaptive algorithms.
Oblivious algorithms are also divided into two
subgroups: deterministic and stochastic
algorithms.
The small size of Network on Chip circuits sets
special requirements for all operations. The
network technology of the Internet is very hard to
straightly shrink to the NoC so the technologies
should be specially adapted to the NoC. The
routing algorithms presented in this report are
difficult to be set in the order of superiority.
Different applications need different routing
algorithms. While some algorithm is suitable to one
system, another algorithm works better in some
other system. However, it can be generalized that in
most of the cases a simple algorithm suits to simple
systems while complex algorithms fit to more
complex systems.
C. Characteristics of NOC’S
With an understanding of NoC and buffer less
NoC design, an important question that remains is:

in what sense do on chip net works differ from
other types of networks? These differences provide
insight into what makes a NoC interesting from a
networking research point of view, and helps to
guide the design of our congestion control
mechanism. We present key properties of both
general and buffer less NoCs.
V. MODULE CLASSIFICATION
The Proposing Methodology having different
classification of modules which are shown below.
They are
1. End to End Data transmission
2. Dynamic Voltage Frequency Scaling
3. Application Traffic Prediction Table
4. Dimension Algorithm
5. Multicore Processor
A. End To End Data Transmission
End to End data transmission is the method of
communication between one end to another end,
Here the end to end data transmission is indicating
the data transmission between one core to another
core. For example every processor using for
different functions, Inside the processor. There are
so many cores for different functions, each core
contains Register, Memory, ALU, MUX, and DMUX,
etc .For this NOC method the core as
interconnected with Each other core and that is for
easy communication. By taking the core with four
terminals that is named as E1, W1, N1 S1.End to
end transmission is the core that is connected with
other core.
A corresponding example for data
communication using EndtoEnd data
transmission is shown in Figure. One can see that
the communication is much simpler in this case.
Rather than checking the packet at each switch
along the path, the packet is just forwarded to its
final destination and a corresponding response is
directly sent back to the source.
B. Dynamic Voltage Frequency Scaling
DVFS gives much more limited energy savings
with relatively high performance overhead as
compared to running workloads at high speed and
then transitioning into low power state Power
consumption has become an overriding concern in
the design of computer systems today. Dynamic
voltage frequency scaling (DVFS) is a power
management technique, that dynamically scales
the voltage and frequency (vf) settings of the cores
so as to provide speed to process the system
workload. Scaling down of voltage levels results in
a quadratic reduction in cores dynamic power
consumption. Many modern cores such as AMD
Opteron and Intel Xeon are equipped with DVFS
capability
Figure 5: Dynamic Voltage Frequency Scaling Block
Diagram
Here dynamic voltage frequency scaling is the
method of monitoring and controlling the
voltage/frequency value of the cores, the above
method that is the end to end data transmission is
monitored by DVFS. Several works have proposed
DVFS in processors for power management.
Recently, work done, it has shown the possibility of
onchip voltage regulators. Prior works have
proposed DVFS for links to manage power in off-
chip networks.
Our proposed schemes are quite different from
all related works discussed here in the sense that
we propose an adaptive frequency router which
adapts to load around its vicinity. We do
distributed throttling of routers by frequency
tuning in on chip networks to manage power and
optimize performance. Moreover, our proposed
schemes manage congestion in the network, and
thereby, give additional throughput gain. To the
best of our knowledge, no prior research has
proposed any technique to tune a router’s
frequency dynamically and adapt to the load
conditions.
C. Application Traffic Prediction Table
It’s a table that having a High and low frequency
range of each core of the system, Depends upon the
ATPT table the DVFS system is giving the voltage to
cores. There are two type of table. However, the
POP uses more memory space than the LVP. The
accuracy of the POP is related to how many
patterns can be tracked in the 2ndlevel table.
Currently, we have not limited the size of the 2nd-
level table, and therefore all the patterns can be
tracked and used to make predictions. But in a
practical implementation, considering the
transistor count overhead, the 2ndlevel table may
be not large enough to hold all the patterns
throughout application execution, so only a subset
of the patterns could be recorded. Consequently,

the accuracy will be dramatically affected by the
replacement policy. When a longer pattern history
is needed, the number of total tracked patterns (the
difference between the POP and the ATPTbased
predictor) will be broadened as Figure 9 shows. To
summarize, the ATPTbased predictor (triangle
dotted line) adaptively selects an LVP in low-
variance traffic, and a POP for highvariance traffic.
Moreover, the experiments show that the ATPT
uses fewer entries than the POP and is suitable for
memory constrained design.
Pattern Learning and Misprediction Recovery:
We give a simple example to demonstrate how the
POP (as a part of the ATPTbased prediction) learns
the communication pattern and recovers from
mispredictionate runtime. shows two different time
periods of the benchmark program execution,
where the POP encountered the same
communication patterns. The first time, the POP
recognized the pattern of core1,and it made an
inaccurate prediction (1 stands for there is a
transmission in that time interval, and 0 stands for
no transmission). The POP will change the
prediction for that communication pattern in the
2ndlevel table when a misprediction is detected.
Afterwards, when the same pattern appears, the
POP can make a correct prediction.
 Last Value Predictor
 Pattern Oriented Predictor
The basic idea is to capture the communication
patterns between parallel tasks, i.e., the end toend
traffic, by using a small table in the network
interface (NI) of each core to record the outgoing
messages from that core. The novel data structure
is called the Applicationdriven Traffic Pattern Table
(ATPT).
Application Traffic Prediction Table: The ATPT
can be applied in the following scenarios:
 Endtoend flow control for controlling the
injection rate.
 Controlling the power mode of switches.
 Modeling the thermal distribution of a chip
while taking both cores and switches in
consideration simultaneously.
 Optimizing the intercommunications of the
application at runtime.
 Three different strategies on DVFS in
communication links are proposed that
satisfy different optimization goals. Inside
the processor each core having some
working frequency range.
D. Dimension Algorithm
Purose of Routing: In many applications, there
are separate nodes of some sort that wish to
communicate with each other using
communications channels of some sort. However,
not all nodes are usually connected to each other,
as connecting all nodes directly to each other
involves lots of wires/cables and/or high-powered
transceivers. Nodes in network must forward other
nodes’ transmissions to correct destination. The
process of determining where to forward packets
and actually doing so is called routing.
Adaptive Routing Algorithm: The issue of inter
processor communication has become paramount
due to the ever increasing speed of each unit and
the amount of Processing Elements (PEs) now
included in mainstream systems. Various
architectures and routing algorithms have
appeared in the last two decades in order to
decrease the overhead to both the data rate as well
as chip size. This document contains an overview of
popular tactics devised for adaptive routing on
multiprocessor microchips. A great variety of
adaptive routing algorithms have been devised for
networking in the more traditional sense. However,
adaptive algorithms for routing in computer
systems with multiple PEs onchip are more recent.
They usually work by introducing flow control
techniques that provide the adaptive behavior
while precluding the possibility of deadlocks. Each
approach may or may not be limited to a specific
topology; both options are explored here.
Deadlocks: Deadlocks in fully adaptive routing can
be a very difficult problem to solve since there are
an infinite amount of possible traffic scenarios.
Node buffer size as well as topology will dictate the
probability of resource deadlock. All solutions
discussed in this document include a form of flow
restrictions, which is to say that only certain
abstract “turns” along the topology are allowed.
Livelocks: A live lock is a type of starvation that
can occur in adaptive routing where misrouting is
permitted. Packets that were routed away from
their destination may become locked in a loop
sequence that persistently reroutes them away
from their goal due to local congestion. The concept
of fairness must be included in an algorithm in
order to preclude this situation. Other solutions
involve the same type of restrictions that are used
for preventing deadlocks.
E. Multicore Processor
DVFS implementations on multicore processors
are more complex than on single core processors
and are often simplified by limiting the available
voltage and/or frequency domains. Chipwide DVFS
forces each core on a package to operate at the
same frequency and voltage. This further

constrains the effectiveness of DVFS because
workloads running on multiple cores must be
analyzed as a whole in order to determine whether
or not to scale frequency.
The most recent AMD Opteron used in our tests
provides a single voltage domain, but independent
frequency domains. Each core can operate at a
different frequency, but the voltage must be no
lower than that required by the core operating at
the highest frequency.
VI. POWER ANALYSIS
A. Power Management
NOCs consist of a set of cores connected with the
communication network. NOC we will be referring
to throughout this paper. The NOC consists of four
cores: MPEG audio, video, speech processing, and
communication core. The cores communicate with
each other via router using a networking protocol.
The design of the networking protocol for NOCs is
beyond the scope of this paper, but has been
addressed in and. In this paper, we enhance each
core’s ability to control its power states by enabling
closed-loop integrated node and network centric
power management with DVS. In order to compute
the power manager’s (PM’s) control, we need to
develop a system model.We model NOC using
Renewal theory as a queuing network with a
number of service points representing cores.
Management of energy consumption under QoS
constraints is formulated as a closed-loop
stochastic control problem. Control theory defines
three different entities in a closed-loop control
system: a system under control, an estimator, and
a controller. PM, as shown in, contains the
controller and the estimator. The estimator
observes the requests coming into the core’s queue,
the state of the core and the incoming power
management requests from the network. Based on
the observations, it tracks any changes in the
system parameters.
Figure 6: Cores Power Management
When a change is detected, it estimates the
new parameter value, and recalculates the power
management control using our fast-optimization
method. The controller implements power
management control calculated by the estimator. It
gives commands to the core that determine its
performance and energy characteristics (frequency
and voltage) in the active state, and chooses when
transition the core into one of the available
low-power states when the core is idle. To illustrate
our power management system operation, we will
use the NOC shown in Fig. 1 with streaming MP3
audio example. Let us assume that initially all
cores in the NOC are in the sleep state. The MP3
core’s controller starts the transition from the sleep
into the active state as soon as the user request
arrives to it via core traffic input. Right at the
beginning of initialization of MP3 decoder, the MP3
core PM also sends a network request via network
PM request port to the communications core to
start its wakeup process.
Thus, any performance penalty that might have
been present without network centric power
management is now masked. Next, the MP3 core
receives encoded MP3 stream data from the
communications core via core traffic input. MP3
estimator watches for the changes in the incoming
data and decoding rates or a discrepancy between
measured and desired MP3 QoS parameters at
runtime. Upon detection, it recalculates the power
management control using our system model and
our fast optimization method and, thus,closes the
optimization loop.
B. Core Parameter
Three main core parameters to estimate
include: core frequency and voltage scaling
characteristics, the time distribution for servicing
incoming core requests, and the characteristics of
the low-power states, such as the transition times
to/from each state. Some core characteristics can
be determined at design time as they depend on
hardware parameters alone, such as the number of
core frequency and voltage settings. Other
parameters need to be tracked at run-time, for
example the properties of service time
distributions. Each NOC core has at least one main
processor. Many of the today’s processor have
multiple active and low-power states. For example,
each of the cores we characterize has one
StrongARM processor, as shown in Fig. 1. The
processor can be configured at runtime by a simple
write to a hardware register to execute at one of
eleven different frequencies.

In addition to the active state, each core can
support multiple lower power states, such as: idle,
sleep, and off. The core enters idle state as soon as
all impeding requests are processed. The
low-power state transitions are controlled by the
PM. The transition time between the active and one
of the low-power states can be best described using
the uniform probability distribution shown in (6)
where and are the minimum and the maximum
transitions times.
C. Power Estimator
The main task of the estimator is to observe the
system behavior and based on that to estimate the
parameters needed for optimization and control.
The quality of power management decisions
strongly depends on estimator’s ability to track
changes of critical parameters at run-time. NOC
power management requires estimation of
workload characteristics, core parameters, and
buffering behavior.
D. Workload Characteristics
Each core’s workload includes request arrivals to
the active state (buffer is not empty), the idle states
(buffer is empty), and the network request arrivals.
We distinguish between active and idle state
arrivals, as they are characterized by two different
distributions, and they also signify a different type
of a decision on the part of the power manager. In
the active state, PM decides only on the appropriate
frequency and voltage setting, where as in the idle
state the primary decision is which low-power state
core, should transition to. The distribution of
network requests only affects the decisions made
by the PM in the idle and the low-power states.
E. Data Coding In Noc Links
The common characteristic of NoC architectures
is that the functional IP blocks communicate with
each other via intelligent switches. The data
communication between IP’s in a NoC takes place
in the form of packets routed through a wormhole
switching mechanism. The packets are broken
down into fixed length flow control units or flits.
The switch blocks need to store only a few flits. The
header flits carry the relevant routing information.
Consequently header decoding enables the
establishment of a path that the subsequent
payload flits simply follow in a pipelined fashion.
There are a few joint crosstalk avoidance and single
error correction codes (CAC/SEC) proposed by
different research groups. Among these joint codes,
the Dual Rail (DR) Code or Duplicate Add Parity
(DAP), Boundary Shift Code (BSC) and Modified
Dual Rail Code (MDR) reduce the switching
capacitance associated with crosstalk from (1+41)
CL to (1+21) CL, where 1 is the ratio of the coupling
capacitance to the bulk capacitance and CL is the
load capacitance, including the self-capacitance of
the wire. However, due to intensive integration and
device shrinkage in the UDSM era, single error
correction will not be sufficient to protect against
different transient malfunctions.
F. Timing Characteristics
The exchange of data among the constituent
blocks in a SoC is becoming an increasingly
difficult task because of growing system size and
nonscalable global wire delay. To cope with these
issues, designers must divide the end-toend
communication medium into multiple pipelined
stages, with the delay in each stage comparable to
the clock-cycle budget. The inter-switch wire
segments, along with the switch blocks, constitute
a highly pipelined communication medium
characterized by link pipelining, deeply pipelined
switches, and latency-insensitive component
design.
VII. ON-CHIP NETWORK PERFORMANCE
ANALYSIS
Network-on-chip provides a structured
communication platform for complex SoC
integration. However, it aggravates the complexity
of on-chip communication design. From the
network perspective, there exists a huge design
space to explore at the network, link and physical
layers. In the network layer, we need to investigate
topology, switching, routing and flow control. In
the link layer, we can examine the impact of link
capacity and link-level flow control schemes on
performance. In the physical layer, we could
inspect wiring, signaling, and robustness issues.
Each of the design considerations (parameters)
also has a number of options to consider. From the
application perspective, the network should not
only be customizable butalso be scalable.
To design an efficient and extensible on-chip
network that suits a specific application or an
application domain, performance analysis is a
crucial and heavy task. The impact of the design
parameters at the different layers and the
performance-cost tradeoffs among these
parameters must be well-understood. The
customization on optimality and extensibility can

sometimes be in conflict with each other. For
instance, a customized irregular topology may be
optimal but not easy to scale. In addition, the
analysis task is very much complicated because of
the un-availability of domain-specific traffic
models. Due to the separation between
computation and communication, a
communication platform may be designed in
parallel with the design of computation. The
concurrent developmentspeeds up time-to-market,
but leaves the development of the communication
platform without sufficiently relevant traffic
knowledge. Therefore we must be able to evaluate
network architectures and analyze their
communication performance with various
communication patterns extensively so as to make
the right design decisions and trade-offs. Once a
network is constructed in hardware, it is difficult,
time-consuming, and expensive to make changes if
performance problems are encountered.
Design decisions include both architecture-level
decisions such as topology, switching, and routing
algorithm, and application-level decisions such as
task-tonode mapping, task scheduling and
synchronization etc.
A. Flow Summary
Figure 7: Flow Summary
Flow summary is nothing but usage of logic
elements in the chip.here we are using 247/6,272
logic element and power consumed by the logic
elements is 4% from the total power. Total number
of register is 117/277. The switching strategy
determines how a message traverses its route.
There are two main switching strategies: circuit
switching and packet switching. Circuit switching
reserves a dedicated end-to-end path from the
source to the destination before starting to
transmit the data. The path can be a real or virtual
circuit.
A packet typically consists of a header, payload
and a tail. The header carries the routing and
sequencing information. The payload is the actual
data to be transmitted. The tail is the end of the
packet and usually contains error-checking code.
Packet-switching can be either connection-oriented
or connection-less. Connection-oriented
communication preserves resources while
connection-less communication does not.
Connection oriented communication can typically
provide a certain degree of commitment for
message delivery bounds. With connection-less
communication, packets are routed individually in
the network in a best-effort manner. The message
delivery is subject to dynamic contention scenarios
in the network, thus is difficult to provide bounds.
However, the network resources can be better
utilized.
B. Network flow control
The network flow control governs how packets
are forwarded in the network, concerning shared
resource allocation and contention resolution. The
shared resources are buffers and links (physical
channels). Essentially a flow control mechanism
deals with the coordination of sending and
receiving packets for the correct delivery of
packets. Due to limited buffers and link
bandwidth, packets may be blocked due to
contention.
Whenever two or more packets attempt to use the
same network resource (e.g., a link or buffer) at the
same time, one of the packets could be stalled in
placed, shunted into buffers, detoured to an
unflavored link, or simply dropped. For
packet-switched networks, there exist buffer less
flow control and buffered flow control .Buffer less
flow control is the simplest form of flow control.
Since there is no buffering in switches, the
resource to be allocated is link bandwidth. It relies
on an arbitration to resolve contentions between
contending packets. After the arbitration, the
winning packet advances over the link. The other
packets are either dropped or misrouted since
there are no buffers.
C. Power Analysis
Figure 8: Power Analysis

 Total Thermal power : 61.54mV
 Dynamic power consumption of cores :
46.13mV
 Static thermal power consumption of core :
36.31mV
 I/O Thermal Power decipation :
15.71mV
Generally speaking, Quality-of-Service (QoS)
defines the level of commitment for packet delivery.
Such a commitment can be correctness of the
result, completion of the transaction, and bounds
on the performance. But, mostly, QoS has a direct
association with bounds in bandwidth, delay and
jitter, since correctness and completion are often
the basic requirements for on-chip message
delivery. Correctness is concerned with packet
integrity (corrupt-less) and packet ordering. It can
be achieved through different means at different
levels. For example, error-correction at the link
layer or re-transmission at the upper layers can be
used to ensure packet integrity.
Figure 9: Multiclock output
Figure 10: Core Output
VIII. CONCLUSION
By solving the power consumption and
increasing the speed of the chip is difficult .here
dynamic voltage/frequency scaling (DVFS) to
adjust the power mode of the switches and links to
match the traffic flows. The challenge is to predict
the traffic flowing through the switches and links in
the next time interval. switch or a link can be
predicted to be idle for a while, then it can be set to
a low power mode or sleep mode by lowering its
voltage or frequency. If the data transmission is
high the DVFS detect the data flow by using ATPT,
Then the data’s are shared to other core of the chip.
By operating this type of functions we can reduce
the power consumption of the chip and increasing
the speed of the system.
APPENDIX
Appendixes, if needed, appear before the
acknowledgment.
ACKNOWLEDGMENT
The preferred spelling of the word
“acknowledgment” in American English is without
an “e” after the “g.” Use the singular heading even if
you have many acknowledgments. Avoid
expressions such as “One of us (S.B.A.) would like
to thank ... .” Instead, write “F. A. Author thanks ...
.” Sponsor and financial support
acknowledgments are placed in the
unnumbered footnote on the first page.
REFERENCES
[1] M. Forsell and S. Kumar, Virtual Distributed Shared
Memory for Network on Chip, Proc. of the 19th
IEEENORCHIP Conference, Nov. 1213, 2015, Kista
[2] YiRan Sun, “Simulation and Performance Evaluation
for Network on Chip”, MSc thesis, Dept. of
Microelectronics and Information Technology, Royal
Institute of Technology, Stockholm. May 2014
[3] S. Bell, B. Edwards, J. Amann, R. Conlin, K. Joyce,
V. Leung,. Tile64 processor: A 64core soc with mesh
interconnect. In SolidState Circuits Conference,
2008.ISSCC 2008. Digest ofTechnical Papers. IEEE
International, pages 88–598, Feb. 2010
[4] L. Shang, L.S.Peh, and N. K. Jha. Dynamic voltage
scaling with links for power optimization of
interconnection networks. In Proc. Ninth
International Symposium on HighPerformance
Computer Architecture HPCA9 2003, pages 91–102,
Feb. 8–12, 2008
[5] Penolazzi S, Jantsch A. A high level power model for
the nostrum noc. In: DSD’06: Proceedings of the 9th
EUROMICRO conference on digital system design;
2006. p. 673
[6] Seo D, Ali A, Lim WT, Rafique N, Thottethodi M.
Nearoptimal worstcase throughput routing for two-
dimensional mesh networks. In: ISCA’05:
Proceedings of the 32nd annual international

symposium on computer architecture; 2005. p.
432–43
[7] Seo D, Ali A, Lim WT, Rafique N, Thottethodi M.
Nearoptimal worstcase throughput routing for two-
dimensional mesh networks. In: ISCA’0 Proceedings
of the 32nd annual international symposium on
computer architecture; 2005. p. 432–43.
[8] Chen G, Chen H, Haurylau M, Nelson N, Albonesi D,
Fauchet PM, et al. Electrical and optical onchip
interconnects in scaled microprocessors. In:
Proceedings of the international symposium on
circuits and systems, vol. 3, 2005. p. 2514–7
[10] Ahmed Hemani, Axel Jantsch, Shashi Kumar,
AdamPostula, Johnny Oberg, Mikael Millberg, and
Dan Lindqvist. Network on chip: architecture for
billion transistor era. In Proceeding of the IEEE
NorChip Conference, November 2003
[11]Kurt Keutzer, Sharad Malik, Richard Newton, Jan
Rabaey, and Alberto Sangiovanni-
Vincentelli.Systemlevel design: Orthogonalization of
concerns and platformbased design. IEEE
Trasnactions on ComputerAided Design of
Integrated Circuits and Systems, 19(12):1523{1543,
Decmber 2000
[12]P. Meloni, S. Carta, R. Argiolas, L. Raffo, F. Angiolini,
Area and power modeling methodologies for
networksonchip, in: Proceedings of 1st International
Conference on NanoNetworks and Workshops, 2000
(NanoNet’06), IEEE Press, New York, September
2000
[13]D. Bertozzi, A. Jalabert, S. Murali, R. Tamhankar, S.
Stergiou, L. Benini, G. De Micheli, Noc synthesis flow
for customized domain specific multiprocessor
systems onchip, IEEE Trans. Parallel Distrib. Syst.
16 (2) (2005) 113–129

IJMTST 2016 | ISSN: 2455-3778 Traffic and Power reduction Routing Algorithm for NOC Cores

More Related Content

What's hot (18)

Viewers also liked (14)

Similar to IJMTST 2016 | ISSN: 2455-3778 Traffic and Power reduction Routing Algorithm for NOC Cores (20)

Recently uploaded (20)

IJMTST 2016 | ISSN: 2455-3778 Traffic and Power reduction Routing Algorithm for NOC Cores