Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology based Network-on-chip

BVICAM’s International Journal of Information Technology
Bharati Vidyapeeth’s Institute of Computer Applications and Management, New Delhi
Performance comparison of XY,OE and DyAd routing algorithm by Load Variation
analysis of 2-Dimensional Mesh topology based Network-on-chip
Parag Parandkar1
,Jayesh kumar Dalal2
,Sumant Katiyal3
1
NIIT University, Neemrana, Dist. Alwar, Rajasthan, India, 2
Samrat Ashok Technological Institute, Vidisha
(M.P.), India, 3
School of Electronics, Devi Ahilya University, Indore (M.P.) India
1
parag.parandkar@niituniversity.in, 2
jayeshdalal7@gmail.com, 3
sumant578@yahoo.com
ABSTRACT
Network on chip is a scalable and flexible communication
architecture for the design of core based System-on-Chip.
Communication performance of a NOC heavily depends on
routing algorithm. XY routing algorithm is distributed
deterministic routing algorithm. Odd-Even (OE) routing
algorithm is distributed adaptive routing algorithm with
deadlock-free ability. DyAD combines the advantages of both
deterministic and adaptive routing schemes. Key metrics which
determines best performance for routing algorithms for
Network-on-Chip architectures are Minimum Latency,
Minimum Power and Maximum Throughput. We demonstrated
the impact of traffic load (bandwidth) variations on average
latency and total network power for three routing algorithms
XY, OE and DyAD on a 3x3 2-dimensional mesh topology. The
simulation is performed on nirgam NoC simulator version 2.1
for constant bit rate traffic condition. The simulation results
reveals the dominance of DyAD over XY and OE algorithms
depicting the minimum values of overall average latency per
channel (in clock cycles per flit) as 1.58871, overall average
latency per channel (in clock cycles per packet) as 9.53226,
overall average latency (in clock cycles per flit) as 26.105, and
total network power as 0.1771 milliwatts, achieved for DyAD
routing algorithm.
KEYWORDS
Network-on-chip; XY routing algorithm; OE routing algorithm;
DyAd routing algorithm.
INTRODUCTION
Network on Chip (NoC) is a new paradigm for System on Chip
(SoC) design [1-5].With the growing complexity and
increasing integration, the commonly used interconnection
techniques for SoC architecture, bus structure, poses practical
physical problems. In NoC paradigm, cores are connected to
each other through a network of routers and they communicate
among themselves through packet-switched communication.
The protocols used in NoC are generally simplified versions of
general communication protocols used in data networks. This
makes it possible to use accepted and mature concepts of
communication networks such as routing algorithms, switching
techniques, flow and congestion control etc. in Network-on-
chip architecture. It allows significant reuse of resources and
provides highly scalable and flexible communication
infrastructure for SoC design.
Data communications between segments of chip are packetized
and transferred through the network. The network consists of
wires and routers. Processors, memories and other IP-blocks
(Intellectual property) are connected to routers. A routing
algorithm plays a significant role on network’s operational
performance.
Fig. 1. 4 X 4 2-Dimensional Mesh NoC
Different routing algorithms are targeted for different
applications. Several routing algorithms need to be investigated
and designed with various features and purposes.
The topic of Network on chip architecture is being introduced
in this section. Section-2 explains about the three basic routing
algorithms namely XY, Odd-even and DyAD routing algorithm
in greater details. Section –3 describe architecture of a 3x3 2-
dimensional mesh topology based NoC. Section –4 discusses
simulation results and analysis of the proposed work. Section-5
ends with conclusion.
XY, OE AND DYAD ROUTING ALGORITHM
The routing algorithm, which defines the path taken by a packet
between the source and the destination, is a main task in
network layer design of NoC. According to where routing
decisions are taken, it is possible to classify the routing as
source and distributed routing [6].
Routing algorithm can be classified on the basis of adaptivity
such as deterministic or adaptive. In deterministic routing, the
path from source to destination is completely determined in
advance by the source and destination address. Examples are
XY routing. In adaptive routing, multiple paths from source to
destinations are possible [7]. There also exits partially adaptive
routing algorithms which restrict certain paths for
communication with deadlock restrictions. Examples are Odd

even routing. They are simple and easy to implement compared
to adaptive routing algorithm. The routing algorithm that uses
shortest path for communication is called minimal routing. The
routing algorithm which uses longer paths for communication
though shorter paths exist is known as non-minimal routing.
Non-minimal routing has some advantages over minimal
routing including possibility of balancing network load and
fault tolerance. In static routing, the path cannot be changed
after a packet leaves the source. In dynamic routing, a path can
be altered anytime depending upon the network conditions.
Routing algorithms can also be defined based on their
implementation: lookup table and Finite State Machine (FSM).
In the following text, three different routing algorithms are
described in details:
XY ROUTING
The XY routing algorithm is one kind of distributed
deterministic routing algorithm. XY routing never runs into
deadlock or livelock [8]. For a 2-Dimesion mesh topology
NoC, each router can be identified by its coordinate (x, y) (Fig.
2). The XY routing algorithm compares the current router
address (Cx,Cy) to the destination router address (Dx,Dy) of
the packet, stored in the header flit [9]. Flits must be routed to
the core port of the router when the (Cx,Cy) address of the
current router is equal to the (Dx,Dy) address.
If this is not the case, the Dx address is firstly compared to the
Cx (horizontal) address. Flits will be routed to the East port
when Cx<Dx, to West when Cx>Dx and if Cx=Dx the header
flit is already horizontally aligned. If this last condition is true,
the Dy (vertical) address is compared to the Cy address. Flits
will be routed to South when Cy<Dy, to North when Cy>Dy. If
the chosen port is busy, the header flit as well as all subsequent
flits of this packet will be blocked. The routing request for this
packet will remain active until a connection is established in
some future execution of the procedure in this router.
The following text is the XY routing algorithm:
/* XY routing Algorithm */
/*Source router: (Sx,Sy);destination router: (Dx,Dy); current
router: (Cx,Cy).*/
begin
if (Dx>Cx) //eastbound messages
return E;
else
if (Dx<Cx) //westbound messages
return W;
else
if (Dx=Cx) { //currently in the same column as
//destination
if (Dy<Cy) //southbound messages
return S;
else
if (Dy>Cy) //northbound messages
return N;
else
if (Dy=Cy) //current router is the destination router
return C;
}
end
The implemention of XY routing algorithm is simple.
However, it is deterministic routing algorithm, which means
this routing algorithm only provides a routing path for a pair of
source and destination. Moreover, XY routing algorithm cannot
avoid from deadlock appearance.
ODD-EVEN ROUTING (OE)
OE routing algorithm is a distributed adaptive routing
algorithm which is based on odd-even turn model [10]. It exerts
some restrictions, for avoiding and preventing from deadlock
appearance. Odd-even turn model facilitates deadlock-free
routing in two-dimensional (2D) meshes with no virtual
channels.
In a two-dimension mesh with dimensions X*Y each node is
identified by its coordinate (x, y) [9]. In this model, a column is
called even if its x dimension element is even numerical
column. Also, a column is called odd if its x dimension element
is an odd number. A turn involves a 90-degree change of
traveling direction. There are eight types of turns, according to
the traveling directions of the associated channels. A turn is
called an ES turn if it involves a change of direction from East
to South. Similarly, we can define the other seven types of
turns, namely EN, WS, WN, SE, SW, NE, and NW turns,
where E, W, S, and N indicate East, West, South, and North,
respectively. As a whole, there are two main theorems in odd-
even algorithm:
Theorem1: No packet is permitted to do EN turn in each node
which is located on an even column. Also, No packet is
permitted to do NW turn in each node that is located on an odd
column.
Theorem 2: No packet is permitted to do ES turn in each node
that is in an even column. Also, no packet is permitted to do
SW turn in each node which is in an odd column.
The following test is a minimal OE routing algorithm in which
avail_dimension_set contains dimensions that are available for
forwarding the packet:
/* OE routing algorithm */
/*Source router: (Sx,Sy);destination router: (Dx,Dy); current
router: (Cx,Cy).*/
begin
avail_dimension_set<-empty;
Ex<-Dx-Cx;
Ey<-Dy-Cy;
if (Ex=0 && Ey=0) //current router is destination
return C;
if (Ex=0){ //current router in same column as destination
if (Ey<0)
add S to avail_dimension_set;
else
add N to avail_dimension_set;
}
else{
if (Ex>0){ //eastbound messages
if (Ey=0){ //current in same row as destination

N
W E
S C
IP
0
(0,0)
1
(0,1)
2
(0,2)
3
(1,0)
4
(1,1)
5
(1,2)
6
(2,0)
7
(2,1)
8
(2,2)
Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology
based Network-on-chip
add E to avail_demision_set;
}
else{
if(Cx % 2 != 0 or Cx=Sx) //N/S turn allowed only in odd
column.
if(Ey < 0)
else
if(Dx% 2 != 0 or Ex != 1) {
//allow to go E only if destination is odd column
add E to avail_dimension_set;
//because N/S turn not allowed in even column
}
}
}
else { // westbound messages
add W to avail_dimension_set;
if(Cx%2=0) //allow to go N/S only in even column, because N-
>W and S->W
//not allowed in odd column
if(Ey<0)
else
}
}
//Select a dimension from avail_dimension_set to forward the
//packet.
End
OE routing algorithm is more complex than XY routing
algorithm. However, it is one kind of adaptive routing
algorithm. For a pair of source and destination, it can provide a
group of routing paths and it can prevent from dead lock
appearance.
DYAD ROUTING
DyAD combines the advantages of both deterministic and
adaptive routing schemes [11]. DyAD is a routing technique
which judiciously switches between deterministic and adaptive
routing based on network congestion’s conditions. Compared
to purely adaptive routers, the overhead of implementing
DyAD is negligible, while the performance is consistently
better.
With DyAD routing each router in the network continuously
monitors its local network load and makes decisions based on
this information. When the network is not congested, a DyAD
router works in a deterministic mode, thus enjoying the low
routing latency enabled by deterministic routing. On the
contrary, when the network becomes congested, the DyAD
router switches back to the adaptive routing mode and thus
avoids the congested links by exploiting other routing paths;
this leads to higher network throughput which is highly
desirable for applications.
The freedom from deadlock and livelock [8] can be guaranteed
when mixing deterministic and adaptive routing modes into the
same NoC.
ARCHITECTURE OF 2 DIMENSION 3X3 MESH
TOPOLOGY NOC
The routing Algorithm is simulated based on a 2-Dimension
3X3 mesh topology NoC (Fig. 2). In the Fig. 2, each circle
represents a tile in the network. Each tile consists of an IP core
connected to a router by a bidirectional core channel (C). A tile
is connected to neighbor tiles by four bidirectional channels (N,
E, S and W). Each tile is identified by a unique integer ID.
Also, each tile can be identified by a pair x-coordinate and y-
coordinate. Our 2-Dimesion 3X3 mesh topology NoC is
designed using wormhole switching mechanism, in which
packets are divided into flits. A packet consists of 3 types of
flits, which are head flit, data flit and tail flit. All the three
routing algorithms, XY routing algorithm, OE routing
algorithm and DyAD routing algorithms are based on these
characteristics.
Fig.2. Architecture of a 2-dimensional 3x3 mesh topology
based NoC
SIMULATION RESULTS AND ANALYSIS
The simulation is performed on NIRGAM simulator, a
simulator for NoC Interconnect Routing and Application
Modeling version 2.1. NIRGAM is an extensible and modular
systemC based simulator [12] as has been depicted in Fig. 3.
Simulations to all the three routing algorithms are performed
under same traffic conditions and simulation control. Tiles are
attached to constant bit rate (CBR) traffic generator. The packet
size is of 20 bytes with random destination mode. The
percentage load, maximum bandwidth to be utilized, is varied
beginning with 10 % to 100 % in the steps of 10 %. The
interval between two successive flits is 2 clock cycles.
Simulation runs for 50000 clock cycles and the clock
frequency is 1 GHz. Synthetic traffic generators generate traffic
in the first 3000 clock cycles with warm-up period of 800 clock
cycles.
Fig. 3 shows the utilization of simulator for the proposed work
elaborating the inputs given to the simulator and outputs taken
from the simulator. There are two bets measures of

Performance of routing algorithms namely, overall average
latency & total network power. The overall average latency in
clock cycles per flit is also measured on a per channel basis on
clock cycles per flit and clock cycles per packet. Total network
power is measured in the units of milliWatts.
Fig.3. Inputs and Outputs to Nirgam NoC Simulator
Table 1 depicts Simulation results for a 3x3 mesh topology
NoC by comparing the impact of Load variation (bandwidth
variation) on overall average latency per channel (in clock
cycles per flit) for XY, OE and DyAd routing algorithms.
Fig.4. shows the graphical representation for simulation data of
Table 1 shows Percentage Load variation vs Overall average
latency per channel (in clock cycles per flit) for OE, XY and
DyAD routing algorithm.
Load
Variation
in %
Overall average latency per channel (in clock
cycles per flit)
OE XY DyAD
10 1.89210 1.90542 1.58871
20 1.77547 2.01144 1.59463
30 1.90437 1.97100 1.59589
40 1.89210 2.02207 1.58785
50 2.07681 2.19148 1.58624
60 2.07850 2.17938 1.60494
70 2.09805 2.09641 1.60256
80 2.13320 2.21263 1.61754
90 2.07522 2.23316 1.58733
100 2.61783 2.69170 1.94854
Table1: Simulation results for load variation verses
Overall average latency per channel (in clock cycles
per flit) for XY, OE and DyAd routing algorithms
Fig. 4. Graph of Percentage Load variation vs Overall average
latency per channel (in clock cycles per flit) for OE, XY
and DyAD routing algorithm
NoC by comparing the impact of Percentage Load variation
(bandwidth variation) on overall average latency per channel
(in clock cycles per packet) for XY, OE and DyAd routing
algorithms.
Fig.5. shows the graphical representation for simulation data of
Table 2 shows Percentage Load variation vs Overall average
latency per channel (in clock cycles per packet) for OE, XY
and DyAD routing algorithm.
Load
Variation
in %
Overall average latency per channel (in clock
cycles per packet)
OE XY DyAD
10 11.35260 11.43250 9.53226
20 10.65280 12.06860 9.56780
30 11.42620 11.82600 9.57534
40 11.35260 12.13240 9.52709
50 12.46080 13.14890 9.51746
60 12.47100 13.07630 9.62963
70 12.58830 12.57840 9.61539
80 12.79920 13.27580 9.70526
90 12.45130 13.39900 9.52396
100 15.70700 16.15020 11.69130
Table 2: Simulation Data results for load variation verses
Overall average latency per channel (in clock cycles per
packet) for XY, OE and DyAd routing algorithms.
Topology
No. of tiles
Destination
Packet Size
Clock Freq.
Traffic Type
Routing Algo
Nirgam
Simulator
Overall Average
Latency
Total Network
Power

Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology
based Network-on-chip
Fig. 5. Graph of Percentage Load variation vs Overall
average latency per channel (in clock cycles per
packet) for OE, XY and DyAD routing algorithm.
(bandwidth variation) on Overall average latency (in clock
cycles per flit) for XY, OE and DyAd routing algorithms.
Fig.6 shows the graphical representation for simulation data of
Table 2 shows Load vs Overall average latency (in clock cycles
per flit) for OE, XY and DyAD routing algorithm.
Load
Variation
in %
Overall average latency
(in clock cycles per flit)
OE XY DyAD
10 30.49320 27.67840 26.10530
20 29.43840 29.04600 25.51350
30 30.77460 29.38890 25.39130
40 31.22220 29.73060 26.19350
50 34.74990 32.62700 27.58700
60 34.25790 32.04840 26.08700
70 34.40660 30.04250 26.25000
80 34.86770 31.81370 25.06520
90 34.46610 32.70150 27.40220
100 49.16410 44.60930 35.58830
Table 3: Simulation results for Percentage Load variation
verses Overall average latency (in clock cycles per flit)
for XY, OE and DyAd routing algorithms
Fig. 6. Graph of Percentage Load variation vs Overall
average latency (in clock cycles per flit) for OE, XY
and DyAD routing algorithm.
(bandwidth variation) on Total Network Power for XY, OE and
DyAd routing algorithms.
Fig.7. shows the graphical representation for data of Table 4
shows Percentage Load variation vs Total Network Power for
OE, XY and DyAD routing algorithm.
Load
Variation
in %
Total Network Power (milliwatts)
OE XY DyAD
10 1.58082 1.44145 0.17711
20 3.16042 2.83479 0.33712
30 3.85637 3.58444 0.41713
40 5.29437 4.86061 0.57989
50 7.96849 7.08161 0.89983
60 7.84587 7.13445 0.84869
70 7.86563 7.06261 0.85435
80 7.81336 7.00110 0.81458
90 7.96460 7.10278 0.89415
100 10.08040 8.54200 1.04783
Table 4: Simulation results for Percentage Load variation
verses Total Network power for XY, OE and DyAd routing
algorithms

Fig. 6. Graph of Percentage Load variation vs Total Network
power (in milliwatts) for OE, XY and DyAD routing
algorithm
CONCLUSION
The routing algorithm is one of network layer researches of a
NoC design, whose design approach can be adapted from a
protocol stack including physical layer, data link layer, network
layer and transport layer. Based on a 2-Dimension 3x3 mesh
topology NoC, three different routing algorithms, XY routing
algorithm, OE routing algorithm and DyAD routing algorithm
are simulated on NIRGAM simulator platform and impact of
Percentage Load variation is compared with four different
parameters namely overall average latency per channel per
packet, overall average latency per channel per flit, overall
average latency per flit and overall network power respectively.
The performance evaluation and the impact of Percentage Load
variation (bandwidth variation) among the routing algorithms
for two important parameters, overall average latency and
overall network power are considered important design criteria
to judge simulator as well as routing algorithm in the NoC
research.
The minimum value of overall average latency per channel (in
clock cycles per flit) is obtained as 1.58871, overall average
latency per channel (in clock cycles per packet) is obtained as
9.53226, overall average latency (in clock cycles per flit) is
obtained as 26.105, and total network power is obtained as
0.1771 milliwatts, achieved for DyAD routing algorithm. Thus
proposed work shows the dominance of DyAD routing
algorithm over OE and XY routing algorithms.
Thus it is concluded that compared to both deterministic and
adaptive routing, significant performance improvements in
terms of total network power as well as overall average latency
can be achieved by using the DyAD approach for constant bit
rate traffic conditions.
FUTURE SCOPE
Our conclusions are just fit for a 2-Dimension 3x3 mesh
topology NoC. For other topologies, as well as taking into
consideration other parameters, additional work needs to be
done in the future.
REFERENCES
[1] P. Guerrier and A. Greiner, “A generic architecture for on-
chip packet-switched interconnections,”Proc. Design,
Automation and Test in Europe Conference and Exhibition
(DATE 2000), Mar. 2000, pp. 250–256, doi:10.1109 /
DATE. 2000.840047.
[2] A. Hemani, A. Jantsch, S. Kumar, A. Postula, J. Oberg, M.
Millberg. et al, “Network on a chip: an architecture for
billion transistor era,” Proc. IEEE NorChip, 2000.
[3] W. Dally and B. Towles, “Route packets, not wires: on-chip
interconnection networks,” Proc. Design Automation
Conference, Jun.2001, pp. 684–689.
[4] M. Sqroi, M. Sheets, A. Mihal, K. Keutzer, S. Malik, J.
Rabaey, et al. “Addressing the system-on-a-chip
interconnect woes through communication-based design”,
Proc. Design Automation Conference, Jun 2001, pp.667-
672.
[5] S. Kumar, A. Jantsch, J. Soininen, M. Forsell, M. Millberg,
J. Oberg. et al, “A network on chip architecture and design
methodology,” Proc. IEEE Computer Society Annual
Symposium on VLSI, Apr. 2002, pp. 105–112, doi:
10.1109/ISVLSI.2002.1016885.
[6] J. Duato, S. Yalamanchili, L. Ni, Interconnection networks:
an engineering approach, Morgan Kaufmann, Revised
Edition, 2002.
[7] L. M. Ni, P. K. Mckinley, “A Survey of Wormhole Routing
Techniques in Direct Networks”, Computer, vol. 26(2), Feb
1993, pp. 62 –76, doi: 10.1109/2.191995.
[8] M. Dehyadgari, M. Nickray, A. Afzali-kusha, Z. Navabi:
Evaluation of Pseudo Adaptive XY Routing Using an
Object OrientedModel for NOC. The 17th International
Conference on Microelectronics, 13–15 December 2005.
[9] Wang Zhang, Ligang Hou, Jinhui Wang, Shuqin Geng,
Wuchen Wu, “Comparison Research between XY and
Odd-Even Routing Algorithm of a 2-Dimension 3X3 Mesh
Topology Network-on-Chip”, WRI Global Congress on
Intelligent Systems, GSIS’09, May 2009, pp. 329-333, doi:
10.1109/GCIS.2009.110.
[10] Ge-Ming Chiu, “The odd-even turn model for adaptive
routing”, IEEE Transactions on parallel and distributed
systems, Jul 2000, pp. 729-738 doi: 10.1109/71.877831.
[11] Jingcao Hu; Marculescu, R., “DyAD - smart routing for
networks-on-chip”, 41st Proceeding of Design automation
conference, San Diego, CA, USA, Jul 2004, pp. 260-263.
[12] L. Jain., “NIRGAM: A Simulator for NoC Interconnect
Routing and Applications Modeling”, date conference,
Sep 2007, pp. 1-2.

Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology based Network-on-chip

More Related Content

What's hot (18)

Viewers also liked (10)

Similar to Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology based Network-on-chip (20)

Performance comparison of XY,OE and DyAd routing algorithm by Load Variation analysis of 2-Dimensional Mesh topology based Network-on-chip