SLIDE 1
Outline
q Intruduction
- 2D mesh NoC
- Routing algorithm
q Background
- Q-routing
- C-routing
q Clustered Q-routing
- Routing table
- Packets format
- Routing algorithm
q Results
2
Optimized Q-learning Model for Distributing Traffic in On-Chip - - PowerPoint PPT Presentation
Optimized Q-learning Model for Distributing Traffic in On-Chip - - PowerPoint PPT Presentation
Optimized Q-learning Model for Distributing Traffic in On-Chip Networks Fahimeh Farahnakian, Masoumeh Ebrahimi ,Masoud Daneshtalab, Pasi Liljeberg, and Juha Plosila University of Turku, Finland Outline q Intruduction 2D mesh NoC
SLIDE 2
SLIDE 3
A Mesh Network on-Chip (NoC) Topology
3
q Each core connect to a router
by a local network interface
q Each router connect to its
neighboring routers through bidirectional links
q Cores communicate with each
- ther using packets
SLIDE 4
Routing Algorithm
- In deterministic routing algorithms a transfer path is
completely determined by the source and destination addresses, like XY .
- In adaptive routing algorithms each packet’s transfer path
determines based on the current network conditions, for example DyXY.
4
Routing policy Deterministic Adaptive
SLIDE 5
5
Motivation(1/3)
An intelligent adaptive routing algorithm which is able to find minimum latency path from a source to a destination using Q-routing.
SLIDE 6
6
Q-routing(1/2)
- Q-routing is an adaptive routing method based on the
Q-learning model in a communication network.
- Each router stores a routing table (Q-table) to maintain
information about the routing cost (Q-value) from itself to the possible destination nodes.
SLIDE 7
qy = waiting time in the packet queue of node y δ =transmission delay over the link from node x to y Q y (z ; d) = the time it would take for node y to send this packet to its Destination via any of node y 's neighbors (z )
7
Q-routing(2/2)
( ) ( ) ( ) ( )
( )
- ld
x y y
- ld
x new x
d y Q q d z Q d y Q d y Q , , , , − + + + = δ γ
s d x y z i j Q x(y,d)
SLIDE 8
C-routing
The C-routing algorithm is a combination of a deterministic routing algorithm (XY) and a Q-routing. Depending on the location of source and destination switches, one of the routing algorithms is invoked.
SLIDE 9
- Each router maintains a Q-table with n×m entries in n×n
2D mesh. The area occupied by the Q-tables:
9
Motivation(2/3)
n : Number of routers in the network m :Number of neighboring routers
SLIDE 10
A clustering approach in order to
- Reduce the area overhead
- Improve the network performance
10
Motivation(3/3)
Clustered Q-routing (CQ-routing)
SLIDE 11
CQ-table
11
D : Number of routers within each cluster.
- A network into C clusters
- CQ-table is maintained for each cluster instead of
each switch.
- The area occupied by the Q-tables:
SLIDE 12
Area Reduction
Mesh Size
- No. of
Clusters
- No. of
Tables AU Q-routing (%) AU C-routing (%) 8×8 16 16 94% 75% 16×16 32 32 98% 83% 32×32 64 64 99% 93% 64×64 128 128 99% 93% 12
SLIDE 13
CQ-routing Algorithm (1/2)
13
Receiving ¡ Cluster ¡ID New ¡ Estimated ¡Latency Destiantion ¡ Cluter ¡ID 2 ¡bits 4 ¡bits 4 ¡bits
Learning packet Data packet
... Upstream Cluster ID QTime 4 bits 2 bits
SLIDE 14
CQ-routing Algorithm (2/2)
14
C1 Cs 3 1 2 8 9
East West North South
10 11
5
- 1
8
- 1
Learning Packet from C1 to C0
New EstimatedLatency
( ) ( )
5 7 5 . 5 5 , 1 − + =
new
Q
(c)
Receiving Cluster_ID Destination Cluster_ID
(a)
East West North South
- 1
4
- 1
10
C3 C12 Cd C5 C4 C6 C7 C11 C10 C9 C8 C13 C14 C2 C1 C12 Cd C5 C4 C6 C7 C11 C10 C9 C8 C13 C14 Cs C3 3 2 10 11 C2 C1 C12 Cd C5 C4 C6 C7 C11 C10 C9 C8 C13 C14 Cs C3 3 2 10 11 C2
East West North South
6
- 1
8
- 1
(b) (a) 15 7 Learning Packet Data Packet 15 1
. . .
SLIDE 15
Results
15
Performance different traffic models in 8×8 2D-mesh
Random Transpose Hotspot
SLIDE 16
Results
16 Random Transpose Hotspot
Performance different traffic models in 14×14 2D-mesh
SLIDE 17