Lecture 12: SMART NoC
|
|
- Jeffry Charles
- 6 years ago
- Views:
Transcription
1 ECE 8823 A / CS ICN Interconnection Networks Spring Lecture 12: SMART NoC Tushar Krishna Assistant Professor School of Electrical and Computer Engineering Georgia Institute of Technology tushar@ece.gatech.edu Krishna et al, Breaking the On-Chip Latcy Barrier Using SMART, HPCA 2013
2 Network Architecture 2 Topology How to connect the nodes ~Road Network Routing Which path should a message take ~Series of road segmts from source to destination Flow Control Wh does the message have to stop/proceed ~Traffic signals at d of each road segmt Router Microarchitecture How to build the routers ~Design of traffic intersection (number of lanes, algorithm for turning red/gre) ICN Spring 2017 L10: Router uarch February 15, 2017
3 Common Pipeline Optimizations 3 BW + RC in parallel Lookahead Routing BW RC VA SA BR ST LT SA + VA in parallel VC Select (switch output port winner selects VC from pool of free VCs) Speculative VA (if VA takes long, speculatively allocate a VC while flit performs SA) (Peh and Dally, HPCA 2001) If SA and VA both successful, go for ST If SA or VA fails, retry next cycle BR + SA in parallel The winner of Input Arbitration is read out and st to the input of the crossbar speculatively Low-load Bypassing Wh no flits in input buffer Speculatively ter ST On port conflict, speculation aborted
4 Express Virtual Channels (ISCA 2007) 4 Analogy Express Trains and Local Trains Flits on Express VCs do not get buffered at intermediate routers Sd a lookahead to ask local flits to wait (i.e., kill switch allocation)
5 Modern Pipelines 5 1-cycle for arbitration (t r ), 1-cycle for traversal (t w ) Used by Tilera s imesh, Intel s Ring, NoC prototypes (Park et al., DAC 2012)
6 Is that the best we can do? 6 T N = (t r +t w ) H + T c + T S 1 t w = 1 Latcy α Hops fundamtal limitation? Can we remove the depdce of latcy on hops? Stay tuned!
7 Designing a 1-cycle network 7 What limits us from designing a 1-cycle network? Is it the wire delay? Repeated global wires can go up to 16mm within 1ns Repeated global wire delay expected to remain constant/decrease slightly Technology = 45nm with technology scaling. Target Clock Period = 1ns Energy'(fJ/bit/mm)' 51# 48# 45# 42# 39# 36# 33# 30# 27# 24# 21# 18# 15# 45nm#(Place4and4Route)# Clocked( 45nm#(Projected*)# Driver( 32nm#(Projected*)# 22nm#(Projected*)# 0# 1# 2# 3# 4# 5# 6# 7# 8# 9# 10# 11# 12# 13# 14# 15# 16# 17# 18# 19# 20# 21# 22# Lgth/period'(mm/ns)' Metal Layer = M6 Repeater Spacing = 1mm Wire Width = DRC min Wire Spacing = 3 DRC min (coupling cap à 0) *DSENT (NOCS 2012): Timing-driv NoC Power Estimation Tool
8 Designing a 1-cycle network 8 What limits us from designing a 1-cycle network? Is it the wire delay? Global repeated wires can transmit up to 10-16mm within 1ns (1GHz) Global On-chip repeated wires wires fast delay ough expected to transmit to remain across fairly the chip constant! Chip within dimsions 1-2 cycles expected at 1GHz to remain ev similar as technology (yield) scales Clock frequcy expected to remain similar (power wall) ~20mm ~20mm ~20mm ~20mm
9 Designing a 1-cycle network 9 What limits us from designing a 1-cycle network? Is it the wire delay? No! Classic scaling challge with wires Wire-delay increases relative to logic delay But Wire-delay in cycles expected to remain constant. Wires fast ough to transmit across chip in 1-2 cycles today and in future.
10 Designing a 1-cycle network 10 What limits us from designing a 1-cycle network? Is it the wire delay? No! Dedicated 1-cycle wire Dream Traversal repeater Actual Traversal HopàStopàHop on-chip router to manage sharing of output link every cycle hop router Shared Links! Number of wires = O(n 2 ) Fully-connected (Impractical) Mesh (Practical) Number of wires = O(n)
11 Designing a 1-cycle network 11 What limits us from designing a 1-cycle network? Is it the wire delay? No! Is it the routers? Yes! 1-cycle repeater Dedicated topology impractical* *unless we design a chip for a specific application router 9-Cycles (Best Case) (single-cycle HopàStopàHop router, no other traffic!) Routers Delay: required 2-4 to cycles. share links Best Case: 1 cycle. Note: this is at every hop Number of wires = O(n 2 ) Fully-connected (Impractical) Mesh (Practical) Number of wires = O(n)
12 Designing a 1-cycle network 12 What limits us from designing a 1-cycle network? Is it the wire delay? No! Is it the routers? Yes! 1-cycle repeater Cycles (Best Case) (single-cycle router, no other traffic!) Dedicated topology impractical* *unless we design a chip for a specific application repeater Routers required to share links Can we get both? Yes! SMART: achieve the performance of dedicated connections over a network of shared links Single-cycle Multi-hop Asynchronous Repeated Traversal 1-cycle (no other traffic)
13 A SMART Network-on-Chip 13 Microarchitecture Bypass path with clockless repeater at each router Flow Control Compete for and reserve a sequce of shared links cycle-by-cycle Dynamically create repeated links ( SMART paths ) betwe any two routers repeater Single-cycle traversal Single-cycle reconfiguration Blue Flow has to stop for Pink flow
14 A SMART Network-on-Chip 14 Microarchitecture Bypass path with clockless repeater at each router Flow Control Compete for and reserve a sequce of shared links cycle-by-cycle Dynamically create repeated links ( SMART paths ) betwe any two routers How well does SMART perform? 88-90% of the performance of an O(n 2 ) wire fully-connected (dream) topology with an O(n) wire SMART NoC Baseline Mesh needs to be clocked 5.4 times faster to match SMART (64-core full-system simulation with real applications) Microarchitecture and Flow Control details next! T. Krishna et al. HPCA 2013 IEEE Computer 2013 IEEE Micro Top Picks 2014 NOCS 2014 (Best Paper Award)
15 Microarchitecture: Data Path 15 network packet Flit Input Links Core North South East West 128-bit Traditional SMART Router Microarchitecture Buffer able (buffer): latch input flit or not Control Bypass Path bypass Queues Input Port local bar Bypass select (): who uses crossbar+link (local or bypass) Output Links Repeater These signals are setup by the control path bar select (): select line for s output
16 Microarchitecture: Control Path 16 Dedicated repeated links from every router to help setup a SMART path Lgth = HPC max hops Width = log 2 (1+HPC max ) bits HPC max (max Hops Per Cycle): maximum number of hops that the underlying wire allows the flit to traverse within a clock cycle e.g., HPC max = 1GHz, 1mm hop Let HPC max = 3 SSR for E direction 2-bit control path R0 R1 R2 R3 R4 128-bit data path
17 SMART Flow Control Divide the packet into flits 17 Assume HPC max = 3 (max Hops Per Cycle) head body tail Divide the route into segmts of HPC max A B HPC max Request a path of desired lgth (in hops) over the SSR wires Intermediate routers arbitrate betwe control requests from various routers and setup buffer, and for the data path No ACK has to be st back! Sd the flit on the data wires May get partial or full SMART path based on conttion that cycle
18 Traversal Example 1: R0 à R3 18 Cycle 1 (Ctrl): R0 sds SSR = 3 (3-hop path request) Assume HPC max = 3 (max Hops Per Cycle) SSR R0 = 3 R0 R1 R2 R3 R4
19 Traversal Example 1: R0 à R3 19 Cycle 1 (Ctrl): R0 sds SSR = 3 (3-hop path request) Assume HPC max = 3 (max Hops Per Cycle) All routers set buffer, and for this request. SSR R0 = 3 R0 R1 R2 buffer 1 R3 R4 local bypass bypass
20 Traversal Example 1: R0 à R3 20 Cycle 1 (Ctrl): R0 sds SSR = 3 (3-hop path request) Assume HPC max = 3 (max Hops Per Cycle) All routers set buffer, and for this request. Cycle 2 (Data): R0 sds flit to R3 SSR R0 = 3 R0 R1 R2 buffer 1 R3 R4 local bypass bypass A SMART path is simply a combination of buffer, and at all the intermediate routers.
21 Example 2: R0 à R3 and R2 à R4 21 Cycle 1 (Ctrl): R0 sds SSR = 3. R2 sds SSR = 2. Challge: only one flit can be st on shared link betwe R2 and R3 at a time SSR R2 = 2 Assume HPC max = 3 (max Hops Per Cycle) SSR R0 = 3 R0 R1 R2 R3 R4 Solution: Routers prioritize betwe path requests based on distance two alternate schemes: Prio=Local and Prio=Bypass
22 Example 2: R0 à R3 and R2 à R4 22 Prio = Local è 0 hop > 1 hop > 2 hop > HPC max hop Cycle 2 (Data): R2 sds flit to R4. R0 s flit blocked at R2. R0 R1 R2 R3 R4 buffer 1 buffer 1 local bypass local bypass Cycle 4+ (Data): R2 sds blocked flit to R3. (after local arbitration + sding ctrl) R0 R1 R2 R3 R4 buffer 1 local
23 Example 2: R0 à R3 and R2 à R4 23 Prio = Bypass è HPC max hop > (HPC max - 1) hop > > 1 hop > 0 hop Cycle 2 (Data): R0 sds flit to R3. R2 s flit waits. R0 R1 R2 R3 R4 buffer 1 local bypass bypass Cycle 3 (Data): R2 sds flit to R4. R0 R1 R2 R3 R4 buffer 1 local bypass
24 Achievable Hops Per Cycle (HPC) 24 At low loads (best case), as good as dedicated wires 6" SMART paths are opportunistic Achievable HPC depds on link conttion Average hops in traffic pattern Average'HPC'(hops)' 5" 4" 3" 2" 1" 0" Fully"Connected"(Dream)" SMART" Baseline"(Mesh)" At high loads (worst case), no worse 0" 20" 40" 60" than 80" the baseline 100" Flit'Injec9on'Rate'(%'of'capacity)'
25 Pipeline and Implemtation 25 SSR R2 SSR R1 SSR R0 SSR R3 min{t ST+LT, T Req + T SA-G } determines critical path (thus HPC max ) Repeater Overhead Energy: Asynchronous repeater consumes 14.3% lower ergy than clocked driver Area: No area overhead since repeaters are embedded in wire-dominated crossbar SA-L SA-G buffer Cycle 0 Switch Allocation Local (SA-L) (can be bypassed at low loads) Cycle 1* R3 SA-G R4 Overhead R5 Energy: ~2% SMART_1D, ~2-6% SMART_2D Area: Cycle < 21% SMART_1D, 1-5% SMART_2D Req + Multi-hop Switch (crossbar) + Switch Allocation Link Traversal (ST+LT) Global (SA-G) (till it is stopped by some buffer=1) HPC max = 13 SMART_1D HPC max = 9 SMART_2D HPC max = 11 *Convtionally (i.e. in baseline), winners of SA-L go for Switch + Link Traversal
26 The Devil is in the Details 26 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in
27 The Devil is in the Details 27 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in
28 Managing Distributed Arbitration 28 Cycle 1: R0 sds SSR = 3. R2 sds SSR = 2. Can differt routers force differt priorities? SSR R0 = 3 SSR R2 = 2 Assume HPC max = 3 (max Hops Per Cycle) R0 R1 R2 R3 R4 Prio = Bypass Prioritize SSR R0 over SSR R2 Prio = Local Prioritize SSR R2 over SSR R0
29 Managing Distributed Arbitration 29 Cycle 1: R0 sds SSR = 3. R2 sds SSR = 2. Can differt routers force differt priorities? No! SSR R0 = 3 SSR R2 = 2 Assume HPC max = 3 (max Hops Per Cycle) R0 R1 R2 R3 buffer 1 R4 local bypass bypass bypass WàE WàE WàE WàE Prio = Bypass Prioritize SSR R0 over SSR R2 R0 s flit incorrectly reaches R4, instead of getting stopped at R3. Prio = Local Prioritize SSR R2 over SSR R0
30 Managing Distributed Arbitration 30 Distributed Conssus: All routers need to take the same decision about multiple contding flits in a distributed manner Solution: All routers follow the same static priority betwe the path setup requests that they receive Prio = Local: 0 hop > 1 hop > (HPC max -1) hop > HPC max hop Prio = Bypass: HPC max hop > (HPC max -1) hop > 1 hop > 0 hop Implication: a router will not receive a flit that it does not expect But can a router not receive a flit that it does expect?
31 Managing Distributed Arbitration 31 Can a router not receive a flit that it does expect? SSR R1 = SSR R0 = 3 Control for N direction Control for E direction R0 R1 R2 R3 R4 Prio = Local
32 Managing Distributed Arbitration 32 Can a router not receive a flit that it does expect? Yes! Is there a performance loss? The R2àR3 link was granted for this cycle, but wt unused. What if some other flit wanted to use it? No. (Prio=Local) R0 buffer 1 R1 R2 buffer 1 R3 R4 local local bypass W->N
33 Impact of False Negatives 33 Cycle 1 (Ctrl Req): R0àR2, R1àR4, R3àR4. SSR R1 = 3 SSR R0 = 2 SSR R3 = 1 Req Priority = Bypass R0 R1 R2 R3 R4 Cycle 2: Flit: R0àR2. forced starvation and throughput loss R0 R1 R2 R3 R4 local bypass buffer 1 local bypass buffer 1
34 Impact of False Negatives 34 Average'flit'latcy'(cycles)' Prio = Bypass increases false negatives at high-loads 28" 24" 20" 16" 12" 8" 4" 0" Prio=Bypass" Prio=Local" Prio = Bypass saturates at ~44-48% injection rate 0" 20" 40" 60" 80" 100" Flit'Injec5on'Rate'(%'of'capacity)'
35 The Devil is in the Details 35 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in
36 Bypass at turns 36 Dest Src
37 Bypass at turns 37 Assume HPC max = 3 Any blue router (HPC max neighborhood) can be reached in 1 cycle Intermediate Ctrl Dest Shortest Path Routing à Any router in shaded HPC max quadrant is pottial intermediate destination Src Separate ctrl path for every possible route. One of the ctrl paths chos during route computation. Only one of these 5 Reqs will be valid and will request for E/N/S output port.
38 Control Arbitration 38 Ctrl Prio = Local Challge: All input and output ports at all participating routers should make consistt decisions simultaneously. Red wins over Blue D B A Solution: 2-level priority among Reqs 1. Distance (i.e. Prio=Local or Prio=Bypass) 2. Direction C local WàE buffer 1 local WàE bypass WàN? S àn? bypass Sà??à E How do we choose betwe Red and Purple?
39 Priority at Output Port 39 At A: Purple wins over Red 2-level Ctrl Req Priority 1. Distance 0 hop > 1 hop > 2 hop (Prio = Local) B 2. Direction Straight hops > Left hops > Right hops 0 hop 1 hop 2 hop 3 hop D E A Intermediate C All gre routers (sources) can request for the North output port at the blue (intermediate) router Assume HPC max = 3 1 Source Routers All routers force same priority (to guarantee no false positives)
40 Priority at Input Port 40 At A: Purple wins over Red At B: Purple wins over Red 0 hop B Intermediate D 1 hop 2 hop 3 hop 4 2 A 3 5 C level Ctrl Req Priority 1. Distance 0 hop > 1 hop > 2 hop (Prio = Local) 2. Direction Straight hops > Left hops > Right hops All gre routers (sources) can request for the South input port at the blue (intermediate) router 1 Source Routers Assume HPC max = 3
41 The Devil is in the Details 41 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
42 The Devil is in the Details 42 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
43 Buffer Managemt 43 R0 à R3, R2 à R4 SSR R2 = 2 SSR R0 = 3 R0 R1 R2 R3 R4 local bypass buffer 1 local bypass buffer 1 How do we guarantee R2 has a free buffer / VC for the flit from R0? Every router has free VC information about its neighbor, just like the baseline. R0 sds only if R1 has a free buffer R1 lets the incoming flit bypass only if R2 has a free buffer, else latches it. Corollary: VCid is allocated after it stops, rather than before starting, since router where it stops is not known. ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
44 The Devil is in the Details 44 Managing Distributed Arbitration Could flits get misrouted? Could flits not arrive wh expected? SMART_2D How can flits bypass routers at turns? Buffer Managemt How is a flit guaranteed a buffer (and in the correct virtual channel) if it is stopped mid-way? How is buffer availability conveyed? Multi-flit packets How does SMART guarantee that flits (head, body, tail) of a packet do not get re-ordered? How do flits know which VC to stop in ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
45 Multi-flit Packets (1) 45 R0 à R3, R2 à R4 SSR R2 = 2 SSR R0 = 3 R0 R1 R2 R3 R4 buffer 1 buffer 1 local bypass local bypass How do we guarantee Body/Tail flits from R0 do not bypass R2 if Head was stopped? If R2 sees a SSR requesting bypass from a router from where it has a buffered flit, it stops all subsequt flits from that router, and buffers them in the same VC ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
46 Multi-flit Packets (2) 46 R0 à R3, R2 à R4 SSR R2 = 2 SSR R0 = 3 R0 R1 R2 R3 R4 buffer 1 buffer 1 local bypass local bypass How do the Body/Tail flits know which VC to stop in if VC is allocated at the router where Head stops? Match using source_id of all the flits of the packet ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
47 Multi-flit Packets (2) 47 R0 à R3, R2 à R4 SSR R2 = 2 SSR R0 = 3 R0 R1 R2 R3 R4 buffer 1 buffer 1 local bypass local bypass What if 2 flits from differt packets from the same source arrive? Virtual Cut-Through ough Buffers to store all flits of a packet all flits of one packet should leave before flits of another packet ICN Spring 2016 L11: Network Interface Tushar Krishna, School of ECE, Georgia Tech February 24, 2016
48 Evaluations 48 Simulation Infrastructure: GEMS (Full-System) + Garnet (NoC) System: 64-core (8x8 Mesh) Technology: 45nm, 1GHz Networks being compared BASELINE (t r =1): Baseline Mesh with 1-cycle router at every hop SMART-<HPC max >_1D: Always stop at turning router Best case delay: 4 cycles (Req à Flit à Req Y à Flit Y ) SMART-<HPC max >_2D: Bypass turning router Best case delay: 2 cycles (Req à Flit) DREAM (T N =1): Conttion-less 1-cycle network (fully-connected)
49 Breaking the latcy barrier 49 Average'flit'latcy'(cycles)' 32" 28" 24" 20" 16" 12" 8" 4" 0" BASELINE"(tr=1)" r SMART:8_1D" SMART:8_2D" DREAM IDEAL"(Tn=1)" N = 1) Lower is better 0" 0.1" 0.2" 0.3" 0.4" 0.5" Injec4on'Rate'(flits/node/cycle)' Average'flit'latcy'(cycles)' 32" 28" 24" 20" 16" 12" 8" 4" 0" BASELINE"(tr=1)" r SMART:8_1D" SMART:8_2D" IDEAL"(Tn=1)" DREAM N = 1) 0" 0.05" 0.1" 0.15" 0.2" 0.25" Injec4on'Rate'(flits/node/cycle)' Uniform Random (Avg Hops = 5.33) Bit Complemt (Avg Hops = 8) SMART reduces low-load latcy to 2-4 cycles across all traffic patterns, indepdt of average hops. Average'flit'latcy'(cycles)' 32" 28" 24" 20" 16" 12" 8" 4" 0" BASELINE"(tr=1)" r SMART:8_1D" SMART:8_2D" DREAM IDEAL"(Tn=1)" N = 1) 0" 0.05" 0.1" 0.15" 0.2" 0.25" Injec4on'Rate'(flits/node/cycle)' Shuffle (Avg Hops = 4) Average'flit'latcy'(cycles)' 32" 28" 24" 20" 16" 12" 8" 4" 0" BASELINE"(tr=1)" r SMART:8_1D" SMART:8_2D" IDEAL"(Tn=1)" DREAM N = 1) 0" 0.05" 0.1" 0.15" 0.2" Injec4on'Rate'(flits/node/cycle)' Transpose (Avg Hops = 6)
50 Impact of HPC max 50 HPC max of 2 and 4 give 1.8 and 3 reduction in latcy Avg$Flit$Latcy$(cycles)$ 32" 28" 24" 20" 16" 12" 8" 4" 0" 0" 0.05" 0.1" 0.15" 0.2" 0.25" Injec4on$Rate$(flits/node/cycle)$ SMART is a better design choice than a baseline 1-cycle router network ev at larger tile size or higher frequcy (i.e. lower HPC max ) HPC max of 8 gives 5.4 reduction in latcy HPC max (max Hops Per Cycle): maximum number of hops that the underlying wire allows the flit to traverse within a clock cycle SMART01_1D" SMART02_1D" SMART04_1D" SMART08_1D" SMART04_2D" SMART08_2D" SMART012_2D" Bit Complemt Traffic A baseline mesh network needs to run at 5.4GHz to match the performance of a 1GHz SMART NoC HPC max will go up as technology scales! (smaller cores, similar frequcy)
51 Full-system Runtime 51 Normalized+Applica/on+Run/me+ 1.2" 1" 0.8" 0.6" 0.4" 0.2" 0" BASELINE"(tr=1)" r SMART58_1D" SMART58_2D" DREAM(T IDEAL"(Tn=1)" N = 1) )" lu" nlu" radix" water5nsq" water5spa9al" blackscholes" canneal" fluidanimate" swap9ons" x264" AVERAGE" SPLASH52" PARSEC" Directory Full5state"Directory" Protocol N Private L2/tile SMART reduces runtime by 26-27% for Private L2 8% off a dream 1-cycle network Normalized+Applica/on+Run/me+ 1.2" 1" 0.8" 0.6" 0.4" 0.2" 0" BASELINE"(tr=1)" SMART58_1D" SMART58_2D" DREAM(T IDEAL"(Tn=1)" N = 1) r )" lu" nlu" radix" water5nsq" water5spa9al" blackscholes" canneal" fluidanimate" swap9ons" x264" AVERAGE" SPLASH52" PARSEC" Directory Full5state"Directory" Protocol N Shared L2/tile SMART reduces runtime by 49-52% for Shared L2 9% off a dream 1-cycle network
52 Backups 52
53 Protocol-level ordering 53 If a virtual network requires protocol-level ordering Only deterministic routing allowed within that virtual network. Priority should be Local > Bypass Guarantees that 2 flits from the same source do not overtake each other at any router.
54 SMART vs. Flatted Butterfly 54 What if we exploit repeated wires and add explicit 1-cyle physical express links? Ports Router Delay SMART 5-port router 1-2 cycle in router [SA-L, SSR+SA-G] Flatted Butterfly 15-port router 3-4 cycle in router [8-14:1 Arbiter, Multi-stage bar] Bisection Bandwidth (BB) 128x8 1-flit req 5-flit resp 128x8 [1x] 7-flit req 35-flit resp 448x8 [3.5x] 2-flit req 10-flit resp 896x8 [7x] 1-flit req 5-flit resp Dyn Power (mw) Area (mm x mm) [~1x] 37 [1.5x] 58 [2.3x] 0.27x x0.47 [3x] 0.53x0.53 [3.9x] 0.7x0.7 [6.7x] R0
55 SMART vs. Flatted Butterfly 55 Assume 1-cycle Flatted Butterfly Router: Highly optimistic assumption Avg$Pkt$Latcy$(cycles)$ 32" 28" 24" 20" 16" 12" 8" 4" 0" 1"pkt"="" 128)bit" 0" 0.1" 0.2" 0.3" 0.4" 0.5" 0.6" 0.7" 0.8" 0.9" 1" SMART always beats FBfly in latcy Injec4on$Rate$(packets/node/cycle)$ FBfly matches SMART in throughput only with 3.5 more wires SMART28_1D" SMART28_2D" FBfly_BB=1x" FBfly_BB=3.5x" FBfly_BB=7x" Better to use SMART and reconfigure 1-cycle multi-hop paths based on traffic rather than use a fixed high-radix topology.
Lecture 3: Flow-Control
High-Performance On-Chip Interconnects for Emerging SoCs http://tusharkrishna.ece.gatech.edu/teaching/nocs_acaces17/ ACACES Summer School 2017 Lecture 3: Flow-Control Tushar Krishna Assistant Professor
More informationLecture 7: Flow Control - I
ECE 8823 A / CS 8803 - ICN Interconnection Networks Spring 2017 http://tusharkrishna.ece.gatech.edu/teaching/icn_s17/ Lecture 7: Flow Control - I Tushar Krishna Assistant Professor School of Electrical
More informationBreaking the On-Chip Latency Barrier Using SMART
Breaking the On-Chip Latency Barrier Using SMART Tushar Krishna Chia-Hsin Owen Chen Woo Cheol Kwon Li-Shiuan Peh Computer Science and Artificial Intelligence Laboratory (CSAIL) Massachusetts Institute
More informationLecture 2: Topology - I
ECE 8823 A / CS 8803 - ICN Interconnection Networks Spring 2017 http://tusharkrishna.ece.gatech.edu/teaching/icn_s17/ Lecture 2: Topology - I Tushar Krishna Assistant Professor School of Electrical and
More informationLecture 3: Topology - II
ECE 8823 A / CS 8803 - ICN Interconnection Networks Spring 2017 http://tusharkrishna.ece.gatech.edu/teaching/icn_s17/ Lecture 3: Topology - II Tushar Krishna Assistant Professor School of Electrical and
More informationOpenSMART: Single-cycle Multi-hop NoC Generator in BSV and Chisel
OpenSMART: Single-cycle Multi-hop NoC Generator in BSV and Chisel Hyoukjun Kwon and Tushar Krishna Georgia Institute of Technology Synergy Lab (http://synergy.ece.gatech.edu) hyoukjun@gatech.edu April
More informationLow-Power Interconnection Networks
Low-Power Interconnection Networks Li-Shiuan Peh Associate Professor EECS, CSAIL & MTL MIT 1 Moore s Law: Double the number of transistors on chip every 2 years 1970: Clock speed: 108kHz No. transistors:
More informationLecture 12: Interconnection Networks. Topics: communication latency, centralized and decentralized switches, routing, deadlocks (Appendix E)
Lecture 12: Interconnection Networks Topics: communication latency, centralized and decentralized switches, routing, deadlocks (Appendix E) 1 Topologies Internet topologies are not very regular they grew
More informationLecture 13: Interconnection Networks. Topics: lots of background, recent innovations for power and performance
Lecture 13: Interconnection Networks Topics: lots of background, recent innovations for power and performance 1 Interconnection Networks Recall: fully connected network, arrays/rings, meshes/tori, trees,
More informationLecture: Interconnection Networks
Lecture: Interconnection Networks Topics: Router microarchitecture, topologies Final exam next Tuesday: same rules as the first midterm 1 Packets/Flits A message is broken into multiple packets (each packet
More informationQuest for High-Performance Bufferless NoCs with Single-Cycle Express Paths and Self-Learning Throttling
Quest for High-Performance Bufferless NoCs with Single-Cycle Express Paths and Self-Learning Throttling Bhavya K. Daya, Li-Shiuan Peh, Anantha P. Chandrakasan Dept. of Electrical Engineering and Computer
More informationLecture 16: On-Chip Networks. Topics: Cache networks, NoC basics
Lecture 16: On-Chip Networks Topics: Cache networks, NoC basics 1 Traditional Networks Huh et al. ICS 05, Beckmann MICRO 04 Example designs for contiguous L2 cache regions 2 Explorations for Optimality
More informationSCORPIO: 36-Core Shared Memory Processor
: 36- Shared Memory Processor Demonstrating Snoopy Coherence on a Mesh Interconnect Chia-Hsin Owen Chen Collaborators: Sunghyun Park, Suvinay Subramanian, Tushar Krishna, Bhavya Daya, Woo Cheol Kwon, Brett
More informationLecture 22: Router Design
Lecture 22: Router Design Papers: Power-Driven Design of Router Microarchitectures in On-Chip Networks, MICRO 03, Princeton A Gracefully Degrading and Energy-Efficient Modular Router Architecture for On-Chip
More informationOpenSMART: An Opensource Singlecycle Multi-hop NoC Generator
OpenSMART: An Opensource Singlecycle Multi-hop NoC Generator Hyoukjun Kwon and Tushar Krishna Georgia Institute of Technology Synergy Lab (http://synergy.ece.gatech.edu) OpenSMART (https://tinyurl.com/get-opensmart)
More informationA Case for Low Frequency Single Cycle Multi Hop NoCs for Energy Efficiency and High Performance
A Case for Low Frequency Single Cycle Multi Hop NoCs for Energy Efficiency and High Performance Monodeep Kar and Tushar Krishna School of Electrical and Computer Engineering Georgia Institute of Technology
More informationThomas Moscibroda Microsoft Research. Onur Mutlu CMU
Thomas Moscibroda Microsoft Research Onur Mutlu CMU CPU+L1 CPU+L1 CPU+L1 CPU+L1 Multi-core Chip Cache -Bank Cache -Bank Cache -Bank Cache -Bank CPU+L1 CPU+L1 CPU+L1 CPU+L1 Accelerator, etc Cache -Bank
More informationJoint consideration of performance, reliability and fault tolerance in regular Networks-on-Chip via multiple spatially-independent interface terminals
Joint consideration of performance, reliability and fault tolerance in regular Networks-on-Chip via multiple spatially-independent interface terminals Philipp Gorski, Tim Wegner, Dirk Timmermann University
More informationLecture 15: PCM, Networks. Today: PCM wrap-up, projects discussion, on-chip networks background
Lecture 15: PCM, Networks Today: PCM wrap-up, projects discussion, on-chip networks background 1 Hard Error Tolerance in PCM PCM cells will eventually fail; important to cause gradual capacity degradation
More informationSynchronized Progress in Interconnection Networks (SPIN) : A new theory for deadlock freedom
ISCA 2018 Session 8B: Interconnection Networks Synchronized Progress in Interconnection Networks (SPIN) : A new theory for deadlock freedom Aniruddh Ramrakhyani Georgia Tech (aniruddh@gatech.edu) Tushar
More informationInterconnection Networks: Topology. Prof. Natalie Enright Jerger
Interconnection Networks: Topology Prof. Natalie Enright Jerger Topology Overview Definition: determines arrangement of channels and nodes in network Analogous to road map Often first step in network design
More informationTopologies. Maurizio Palesi. Maurizio Palesi 1
Topologies Maurizio Palesi Maurizio Palesi 1 Network Topology Static arrangement of channels and nodes in an interconnection network The roads over which packets travel Topology chosen based on cost and
More informationEECS 570. Lecture 19 Interconnects: Flow Control. Winter 2018 Subhankar Pal
Lecture 19 Interconnects: Flow Control Winter 2018 Subhankar Pal http://www.eecs.umich.edu/courses/eecs570/ Slides developed in part by Profs. Adve, Falsafi, Hill, Lebeck, Martin, Narayanasamy, Nowatzyk,
More informationPacket Switch Architecture
Packet Switch Architecture 3. Output Queueing Architectures 4. Input Queueing Architectures 5. Switching Fabrics 6. Flow and Congestion Control in Sw. Fabrics 7. Output Scheduling for QoS Guarantees 8.
More informationPacket Switch Architecture
Packet Switch Architecture 3. Output Queueing Architectures 4. Input Queueing Architectures 5. Switching Fabrics 6. Flow and Congestion Control in Sw. Fabrics 7. Output Scheduling for QoS Guarantees 8.
More informationSwitching/Flow Control Overview. Interconnection Networks: Flow Control and Microarchitecture. Packets. Switching.
Switching/Flow Control Overview Interconnection Networks: Flow Control and Microarchitecture Topology: determines connectivity of network Routing: determines paths through network Flow Control: determine
More informationBasic Low Level Concepts
Course Outline Basic Low Level Concepts Case Studies Operation through multiple switches: Topologies & Routing v Direct, indirect, regular, irregular Formal models and analysis for deadlock and livelock
More informationPrediction Router: Yet another low-latency on-chip router architecture
Prediction Router: Yet another low-latency on-chip router architecture Hiroki Matsutani Michihiro Koibuchi Hideharu Amano Tsutomu Yoshinaga (Keio Univ., Japan) (NII, Japan) (Keio Univ., Japan) (UEC, Japan)
More informationInterconnection Networks
Lecture 18: Interconnection Networks Parallel Computer Architecture and Programming CMU 15-418/15-618, Spring 2015 Credit: many of these slides were created by Michael Papamichael This lecture is partially
More informationEECS 598: Integrating Emerging Technologies with Computer Architecture. Lecture 12: On-Chip Interconnects
1 EECS 598: Integrating Emerging Technologies with Computer Architecture Lecture 12: On-Chip Interconnects Instructor: Ron Dreslinski Winter 216 1 1 Announcements Upcoming lecture schedule Today: On-chip
More informationPseudo-Circuit: Accelerating Communication for On-Chip Interconnection Networks
Department of Computer Science and Engineering, Texas A&M University Technical eport #2010-3-1 seudo-circuit: Accelerating Communication for On-Chip Interconnection Networks Minseon Ahn, Eun Jung Kim Department
More informationLecture 13: System Interface
ECE 8823 A / CS 8803 - ICN Interconnection Networks Spring 2017 http://tusharkrishna.ece.gatech.edu/teaching/icn_s17/ Lecture 13: System Interface Tushar Krishna Assistant Professor School of Electrical
More informationTDT Appendix E Interconnection Networks
TDT 4260 Appendix E Interconnection Networks Review Advantages of a snooping coherency protocol? Disadvantages of a snooping coherency protocol? Advantages of a directory coherency protocol? Disadvantages
More informationA thesis presented to. the faculty of. In partial fulfillment. of the requirements for the degree. Master of Science. Yixuan Zhang.
High-Performance Crossbar Designs for Network-on-Chips (NoCs) A thesis presented to the faculty of the Russ College of Engineering and Technology of Ohio University In partial fulfillment of the requirements
More informationLecture 15: NoC Innovations. Today: power and performance innovations for NoCs
Lecture 15: NoC Innovations Today: power and performance innovations for NoCs 1 Network Power Power-Driven Design of Router Microarchitectures in On-Chip Networks, MICRO 03, Princeton Energy for a flit
More informationModule 17: "Interconnection Networks" Lecture 37: "Introduction to Routers" Interconnection Networks. Fundamentals. Latency and bandwidth
Interconnection Networks Fundamentals Latency and bandwidth Router architecture Coherence protocol and routing [From Chapter 10 of Culler, Singh, Gupta] file:///e /parallel_com_arch/lecture37/37_1.htm[6/13/2012
More informationMinBD: Minimally-Buffered Deflection Routing for Energy-Efficient Interconnect
MinBD: Minimally-Buffered Deflection Routing for Energy-Efficient Interconnect Chris Fallin, Greg Nazario, Xiangyao Yu*, Kevin Chang, Rachata Ausavarungnirun, Onur Mutlu Carnegie Mellon University *CMU
More informationLecture 18: Communication Models and Architectures: Interconnection Networks
Design & Co-design of Embedded Systems Lecture 18: Communication Models and Architectures: Interconnection Networks Sharif University of Technology Computer Engineering g Dept. Winter-Spring 2008 Mehdi
More informationPower and Performance Efficient Partial Circuits in Packet-Switched Networks-on-Chip
2013 21st Euromicro International Conference on Parallel, Distributed, and Network-Based Processing Power and Performance Efficient Partial Circuits in Packet-Switched Networks-on-Chip Nasibeh Teimouri
More informationLecture 25: Interconnection Networks, Disks. Topics: flow control, router microarchitecture, RAID
Lecture 25: Interconnection Networks, Disks Topics: flow control, router microarchitecture, RAID 1 Virtual Channel Flow Control Each switch has multiple virtual channels per phys. channel Each virtual
More informationDynamic Packet Fragmentation for Increased Virtual Channel Utilization in On-Chip Routers
Dynamic Packet Fragmentation for Increased Virtual Channel Utilization in On-Chip Routers Young Hoon Kang, Taek-Jun Kwon, and Jeff Draper {youngkan, tjkwon, draper}@isi.edu University of Southern California
More informationLecture 24: Interconnection Networks. Topics: topologies, routing, deadlocks, flow control
Lecture 24: Interconnection Networks Topics: topologies, routing, deadlocks, flow control 1 Topology Examples Grid Torus Hypercube Criteria Bus Ring 2Dtorus 6-cube Fully connected Performance Bisection
More informationEfficient Throughput-Guarantees for Latency-Sensitive Networks-On-Chip
ASP-DAC 2010 20 Jan 2010 Session 6C Efficient Throughput-Guarantees for Latency-Sensitive Networks-On-Chip Jonas Diemer, Rolf Ernst TU Braunschweig, Germany diemer@ida.ing.tu-bs.de Michael Kauschke Intel,
More informationTopologies. Maurizio Palesi. Maurizio Palesi 1
Topologies Maurizio Palesi Maurizio Palesi 1 Network Topology Static arrangement of channels and nodes in an interconnection network The roads over which packets travel Topology chosen based on cost and
More informationNetwork-on-chip (NOC) Topologies
Network-on-chip (NOC) Topologies 1 Network Topology Static arrangement of channels and nodes in an interconnection network The roads over which packets travel Topology chosen based on cost and performance
More informationLecture 23: Router Design
Lecture 23: Router Design Papers: A Gracefully Degrading and Energy-Efficient Modular Router Architecture for On-Chip Networks, ISCA 06, Penn-State ViChaR: A Dynamic Virtual Channel Regulator for Network-on-Chip
More informationECE 669 Parallel Computer Architecture
ECE 669 Parallel Computer Architecture Lecture 21 Routing Outline Routing Switch Design Flow Control Case Studies Routing Routing algorithm determines which of the possible paths are used as routes how
More informationAchieving Lightweight Multicast in Asynchronous Networks-on-Chip Using Local Speculation
Achieving Lightweight Multicast in Asynchronous Networks-on-Chip Using Local Speculation Kshitij Bhardwaj Dept. of Computer Science Columbia University Steven M. Nowick 2016 ACM/IEEE Design Automation
More informationLecture 12: Interconnection Networks. Topics: dimension/arity, routing, deadlock, flow control
Lecture 12: Interconnection Networks Topics: dimension/arity, routing, deadlock, flow control 1 Interconnection Networks Recall: fully connected network, arrays/rings, meshes/tori, trees, butterflies,
More informationExpress Virtual Channels: Towards the Ideal Interconnection Fabric
Express Virtual Channels: Towards the Ideal Interconnection Fabric Amit Kumar, Li-Shiuan Peh, Partha Kundu and Niraj K. Jha Dept. of Electrical Engineering, Princeton University, Princeton, NJ 8544 Microprocessor
More informationA Dynamic NOC Arbitration Technique using Combination of VCT and XY Routing
727 A Dynamic NOC Arbitration Technique using Combination of VCT and XY Routing 1 Bharati B. Sayankar, 2 Pankaj Agrawal 1 Electronics Department, Rashtrasant Tukdoji Maharaj Nagpur University, G.H. Raisoni
More informationExtending the Performance of Hybrid NoCs beyond the Limitations of Network Heterogeneity
Journal of Low Power Electronics and Applications Article Extending the Performance of Hybrid NoCs beyond the Limitations of Network Heterogeneity Michael Opoku Agyeman 1, *, Wen Zong 2, Alex Yakovlev
More informationNEtwork-on-Chip (NoC) [3], [6] is a scalable interconnect
1 A Soft Tolerant Network-on-Chip Router Pipeline for Multi-core Systems Pavan Poluri and Ahmed Louri Department of Electrical and Computer Engineering, University of Arizona Email: pavanp@email.arizona.edu,
More informationLecture 26: Interconnects. James C. Hoe Department of ECE Carnegie Mellon University
18 447 Lecture 26: Interconnects James C. Hoe Department of ECE Carnegie Mellon University 18 447 S18 L26 S1, James C. Hoe, CMU/ECE/CALCM, 2018 Housekeeping Your goal today get an overview of parallel
More informationThe Design and Implementation of a Low-Latency On-Chip Network
The Design and Implementation of a Low-Latency On-Chip Network Robert Mullins 11 th Asia and South Pacific Design Automation Conference (ASP-DAC), Jan 24-27 th, 2006, Yokohama, Japan. Introduction Current
More informationCS 258, Spring 99 David E. Culler Computer Science Division U.C. Berkeley Wide links, smaller routing delay Tremendous variation 3/19/99 CS258 S99 2
Real Machines Interconnection Network Topology Design Trade-offs CS 258, Spring 99 David E. Culler Computer Science Division U.C. Berkeley Wide links, smaller routing delay Tremendous variation 3/19/99
More informationLecture: Transactional Memory, Networks. Topics: TM implementations, on-chip networks
Lecture: Transactional Memory, Networks Topics: TM implementations, on-chip networks 1 Summary of TM Benefits As easy to program as coarse-grain locks Performance similar to fine-grain locks Avoids deadlock
More informationLecture: Interconnection Networks. Topics: TM wrap-up, routing, deadlock, flow control, virtual channels
Lecture: Interconnection Networks Topics: TM wrap-up, routing, deadlock, flow control, virtual channels 1 TM wrap-up Eager versioning: create a log of old values Handling problematic situations with a
More informationLecture 14: Large Cache Design III. Topics: Replacement policies, associativity, cache networks, networking basics
Lecture 14: Large Cache Design III Topics: Replacement policies, associativity, cache networks, networking basics 1 LIN Qureshi et al., ISCA 06 Memory level parallelism (MLP): number of misses that simultaneously
More informationOASIS NoC Architecture Design in Verilog HDL Technical Report: TR OASIS
OASIS NoC Architecture Design in Verilog HDL Technical Report: TR-062010-OASIS Written by Kenichi Mori ASL-Ben Abdallah Group Graduate School of Computer Science and Engineering The University of Aizu
More informationISSN Vol.03,Issue.06, August-2015, Pages:
WWW.IJITECH.ORG ISSN 2321-8665 Vol.03,Issue.06, August-2015, Pages:0920-0924 Performance and Evaluation of Loopback Virtual Channel Router with Heterogeneous Router for On Chip Network M. VINAY KRISHNA
More informationSwizzle Switch: A Self-Arbitrating High-Radix Crossbar for NoC Systems
1 Swizzle Switch: A Self-Arbitrating High-Radix Crossbar for NoC Systems Ronald Dreslinski, Korey Sewell, Thomas Manville, Sudhir Satpathy, Nathaniel Pinckney, Geoff Blake, Michael Cieslak, Reetuparna
More informationA VERIOG-HDL IMPLEMENTATION OF VIRTUAL CHANNELS IN A NETWORK-ON-CHIP ROUTER. A Thesis SUNGHO PARK
A VERIOG-HDL IMPLEMENTATION OF VIRTUAL CHANNELS IN A NETWORK-ON-CHIP ROUTER A Thesis by SUNGHO PARK Submitted to the Office of Graduate Studies of Texas A&M University in partial fulfillment of the requirements
More informationES1 An Introduction to On-chip Networks
December 17th, 2015 ES1 An Introduction to On-chip Networks Davide Zoni PhD mail: davide.zoni@polimi.it webpage: home.dei.polimi.it/zoni Sources Main Reference Book (for the examination) Designing Network-on-Chip
More informationThe Runahead Network-On-Chip
The Network-On-Chip Zimo Li University of Toronto zimo.li@mail.utoronto.ca Joshua San Miguel University of Toronto joshua.sanmiguel@mail.utoronto.ca Natalie Enright Jerger University of Toronto enright@ece.utoronto.ca
More informationFall 2012 Parallel Computer Architecture Lecture 21: Interconnects IV. Prof. Onur Mutlu Carnegie Mellon University 10/29/2012
18-742 Fall 2012 Parallel Computer Architecture Lecture 21: Interconnects IV Prof. Onur Mutlu Carnegie Mellon University 10/29/2012 New Review Assignments Were Due: Sunday, October 28, 11:59pm. Das et
More informationNOW Handout Page 1. Outline. Networks: Routing and Design. Routing. Routing Mechanism. Routing Mechanism (cont) Properties of Routing Algorithms
Outline Networks: Routing and Design Routing Switch Design Case Studies CS 5, Spring 99 David E. Culler Computer Science Division U.C. Berkeley 3/3/99 CS5 S99 Routing Recall: routing algorithm determines
More information18-740/640 Computer Architecture Lecture 16: Interconnection Networks. Prof. Onur Mutlu Carnegie Mellon University Fall 2015, 11/4/2015
18-740/640 Computer Architecture Lecture 16: Interconnection Networks Prof. Onur Mutlu Carnegie Mellon University Fall 2015, 11/4/2015 Required Readings Required Reading Assignment: Dubois, Annavaram,
More informationInterconnection Networks: Flow Control. Prof. Natalie Enright Jerger
Interconnection Networks: Flow Control Prof. Natalie Enright Jerger Switching/Flow Control Overview Topology: determines connectivity of network Routing: determines paths through network Flow Control:
More informationFlow Control can be viewed as a problem of
NOC Flow Control 1 Flow Control Flow Control determines how the resources of a network, such as channel bandwidth and buffer capacity are allocated to packets traversing a network Goal is to use resources
More informationInterconnection Networks: Routing. Prof. Natalie Enright Jerger
Interconnection Networks: Routing Prof. Natalie Enright Jerger Routing Overview Discussion of topologies assumed ideal routing In practice Routing algorithms are not ideal Goal: distribute traffic evenly
More informationCONGESTION AWARE ADAPTIVE ROUTING FOR NETWORK-ON-CHIP COMMUNICATION. Stephen Chui Bachelor of Engineering Ryerson University, 2012.
CONGESTION AWARE ADAPTIVE ROUTING FOR NETWORK-ON-CHIP COMMUNICATION by Stephen Chui Bachelor of Engineering Ryerson University, 2012 A thesis presented to Ryerson University in partial fulfillment of the
More informationNETWORK-ON-CHIPS (NoCs) [1], [2] design paradigm
IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS 1 Extending the Energy Efficiency and Performance With Channel Buffers, Crossbars, and Topology Analysis for Network-on-Chips Dominic DiTomaso,
More informationInterconnection Networks
Lecture 17: Interconnection Networks Parallel Computer Architecture and Programming A comment on web site comments It is okay to make a comment on a slide/topic that has already been commented on. In fact
More informationQuality-of-Service for a High-Radix Switch
Quality-of-Service for a High-Radix Switch Nilmini Abeyratne, Supreet Jeloka, Yiping Kang, David Blaauw, Ronald G. Dreslinski, Reetuparna Das, and Trevor Mudge University of Michigan 51 st DAC 06/05/2014
More informationECE 4750 Computer Architecture, Fall 2017 T06 Fundamental Network Concepts
ECE 4750 Computer Architecture, Fall 2017 T06 Fundamental Network Concepts School of Electrical and Computer Engineering Cornell University revision: 2017-10-17-12-26 1 Network/Roadway Analogy 3 1.1. Running
More informationAsynchronous Bypass Channel Routers
1 Asynchronous Bypass Channel Routers Tushar N. K. Jain, Paul V. Gratz, Alex Sprintson, Gwan Choi Department of Electrical and Computer Engineering, Texas A&M University {tnj07,pgratz,spalex,gchoi}@tamu.edu
More information3D WiNoC Architectures
Interconnect Enhances Architecture: Evolution of Wireless NoC from Planar to 3D 3D WiNoC Architectures Hiroki Matsutani Keio University, Japan Sep 18th, 2014 Hiroki Matsutani, "3D WiNoC Architectures",
More informationRouting Algorithms. Review
Routing Algorithms Today s topics: Deterministic, Oblivious Adaptive, & Adaptive models Problems: efficiency livelock deadlock 1 CS6810 Review Network properties are a combination topology topology dependent
More informationPRIORITY BASED SWITCH ALLOCATOR IN ADAPTIVE PHYSICAL CHANNEL REGULATOR FOR ON CHIP INTERCONNECTS. A Thesis SONALI MAHAPATRA
PRIORITY BASED SWITCH ALLOCATOR IN ADAPTIVE PHYSICAL CHANNEL REGULATOR FOR ON CHIP INTERCONNECTS A Thesis by SONALI MAHAPATRA Submitted to the Office of Graduate and Professional Studies of Texas A&M University
More informationCS252 Graduate Computer Architecture Lecture 14. Multiprocessor Networks March 9 th, 2011
CS252 Graduate Computer Architecture Lecture 14 Multiprocessor Networks March 9 th, 2011 John Kubiatowicz Electrical Engineering and Computer Sciences University of California, Berkeley http://www.eecs.berkeley.edu/~kubitron/cs252
More informationEvaluating Bufferless Flow Control for On-Chip Networks
Evaluating Bufferless Flow Control for On-Chip Networks George Michelogiannakis, Daniel Sanchez, William J. Dally, Christos Kozyrakis Stanford University In a nutshell Many researchers report high buffer
More informationPhastlane: A Rapid Transit Optical Routing Network
Phastlane: A Rapid Transit Optical Routing Network Mark Cianchetti, Joseph Kerekes, and David Albonesi Computer Systems Laboratory Cornell University The Interconnect Bottleneck Future processors: tens
More informationNoC Test-Chip Project: Working Document
NoC Test-Chip Project: Working Document Michele Petracca, Omar Ahmad, Young Jin Yoon, Frank Zovko, Luca Carloni and Kenneth Shepard I. INTRODUCTION This document describes the low-power high-performance
More informationLocality-Oblivious Cache Organization leveraging Single-Cycle Multi-Hop NoCs
Locality-Oblivious Cache Organization leveraging Single-Cycle Multi-Hop NoCs Woo-Cheol Kwon Department of EECS, MIT Cambridge, MA 9 wckwon@csail.mit.edu Tushar Krishna Intel Corporation, VSSAD Hudson,
More informationSCORPIO: A 36-Core Research Chip Demonstrating Snoopy Coherence on a Scalable Mesh NoC with In- Network Ordering
SCORPIO: A 36-Core Research Chip Demonstrating Snoopy Coherence on a Scalable Mesh NoC with In- Network Ordering The MIT Faculty has made this article openly available. Please share how this access benefits
More information548 IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 30, NO. 4, APRIL 2011
548 IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 30, NO. 4, APRIL 2011 Extending the Effective Throughput of NoCs with Distributed Shared-Buffer Routers Rohit Sunkam
More informationNoCAlert: An On-Line and Real- Time Fault Detection Mechanism for Network-on-Chip Architectures
NoCAlert: An On-Line and Real- Time Fault Detection Mechanism for Network-on-Chip Architectures Andreas Prodromou, Andreas Panteli, Chrysostomos Nicopoulos, and Yiannakis Sazeides University of Cyprus
More informationDesign of a high-throughput distributed shared-buffer NoC router
Design of a high-throughput distributed shared-buffer NoC router The MIT Faculty has made this article openly available Please share how this access benefits you Your story matters Citation As Published
More informationA Survey of Techniques for Power Aware On-Chip Networks.
A Survey of Techniques for Power Aware On-Chip Networks. Samir Chopra Ji Young Park May 2, 2005 1. Introduction On-chip networks have been proposed as a solution for challenges from process technology
More informationA Low Latency Router Supporting Adaptivity for On-Chip Interconnects
A Low Latency Supporting Adaptivity for On-Chip Interconnects 34.2 Jongman Kim, Dongkook Park, T. Theocharides, N. Vijaykrishnan and Chita R. Das Department of Computer Science and Engineering The Pennsylvania
More informationA Novel Energy Efficient Source Routing for Mesh NoCs
2014 Fourth International Conference on Advances in Computing and Communications A ovel Energy Efficient Source Routing for Mesh ocs Meril Rani John, Reenu James, John Jose, Elizabeth Isaac, Jobin K. Antony
More informationVLPW: THE VERY LONG PACKET WINDOW ARCHITECTURE FOR HIGH THROUGHPUT NETWORK-ON-CHIP ROUTER DESIGNS
VLPW: THE VERY LONG PACKET WINDOW ARCHITECTURE FOR HIGH THROUGHPUT NETWORK-ON-CHIP ROUTER DESIGNS A Thesis by HAIYIN GU Submitted to the Office of Graduate Studies of Texas A&M University in partial fulfillment
More informationRouting Algorithm. How do I know where a packet should go? Topology does NOT determine routing (e.g., many paths through torus)
Routing Algorithm How do I know where a packet should go? Topology does NOT determine routing (e.g., many paths through torus) Many routing algorithms exist 1) Arithmetic 2) Source-based 3) Table lookup
More informationRecall: The Routing problem: Local decisions. Recall: Multidimensional Meshes and Tori. Properties of Routing Algorithms
CS252 Graduate Computer Architecture Lecture 16 Multiprocessor Networks (con t) March 14 th, 212 John Kubiatowicz Electrical Engineering and Computer Sciences University of California, Berkeley http://www.eecs.berkeley.edu/~kubitron/cs252
More informationEE 6900: Interconnection Networks for HPC Systems Fall 2016
EE 6900: Interconnection Networks for HPC Systems Fall 2016 Avinash Karanth Kodi School of Electrical Engineering and Computer Science Ohio University Athens, OH 45701 Email: kodi@ohio.edu 1 Acknowledgement:
More informationLatency Criticality Aware On-Chip Communication
Latency Criticality Aware On-Chip Communication Zheng Li, Jie Wu, Li Shang, Robert P. Dick, and Yihe Sun Tsinghua National Laboratory for Information Science and Technology, Inst. of Microelectronics,
More informationDLABS: a Dual-Lane Buffer-Sharing Router Architecture for Networks on Chip
DLABS: a Dual-Lane Buffer-Sharing Router Architecture for Networks on Chip Anh T. Tran and Bevan M. Baas Department of Electrical and Computer Engineering University of California - Davis, USA {anhtr,
More informationInterconnection Network
Interconnection Network Recap: Generic Parallel Architecture A generic modern multiprocessor Network Mem Communication assist (CA) $ P Node: processor(s), memory system, plus communication assist Network
More information4. Networks. in parallel computers. Advances in Computer Architecture
4. Networks in parallel computers Advances in Computer Architecture System architectures for parallel computers Control organization Single Instruction stream Multiple Data stream (SIMD) All processors
More information