udirec: Unified Diagnosis and Reconfiguration for Frugal Bypass of NoC Faults

Transcription

1 1/45 1/22 MICRO-46, 9 th December- 213 Davis, California udirec: Unified Diagnosis and Reconfiguration for Frugal Bypass of NoC Faults Ritesh Parikh and Valeria Bertacco Electrical Engineering & Computer Science Department The University of Michigan, Ann Arbor

2 Device Scaling and Processor Evolution 65nm 45nm Shrinking transistor size Transition to multicore 32nm 22nm Waning reliability NoC based interconnect Simple bus Point-to-point bus High-speed ring network 2D mesh network-on-chip Core2Duo - 27 Intel Nehalem - 28 Intel Sandy Bridge Intel SCC- 29 2/45 2/22

3 Device Scaling and Processor Evolution 65nm Growing adoption of Network-on-Chip: 45nm 32nm 22nm Sole medium of on-chip communication Waning reliability Potential single-point of failure NoC based interconnect Shrinking transistor size Transition to multicore Simple bus Intel Identifies Cougar Point Error, Point-to-point SATA bus Connection High-speed Single ring network Point of 2D Failure. mesh network-on-chip January, 211. Cost: $7 Million Core2Duo - 27 Intel Nehalem - 28 Intel Sandy Bridge Intel SCC- 29 3/45 3/22

4 On-Chip Interconnect Evolution core/ip core-1 network interface core-2 network interface bus core-3 network interface router #1 core-4 network interface router #2 link Bus-based multicore/soc T data B packet B B H NoC-based multicore/soc Routers connected via links Cores connected via network interface (NI) tail T B B B H body header i n p u t s input port FIFO routing router #3 arbiters /allocators router #4 output port o u t p u t s flits crossbar credit, FIFO and routing control 4/45 4/22

5 5/45 5/22 Permanent Faults in NoCs Permanent wear-out Device aging NoCs: significant silicon footprint and heavy activity Detection Diagnosis Reconfiguration Recovery Detect if fault has occurred proc NI rtr S Diagnose where fault has occurred D Reconfigure network to account for fault cannot send on faulty path need to re-route around fault Recover and resume normal operation udirec solves two problems: Fault diagnosis at a fine resolution Reconfiguration to find new routes

6 6/45 6/22 Contributions udirec (for unified DIagnosis and REConfiguration) incorporates: routing-aware scheme for diagnosis route-reconfiguration to circumvent faults Fine-grain fault model for NoCs derived from: end-to-end diagnosis scheme frugal reconfiguration algorithm A deadlock-free routing algorithm for irregular networks with unidirectional links S D Error!! Error!! Error!! Tightly integrated software implementation enables: low overhead no performance overhead during fault-free execution

7 7/45 7/22 Rest of This Talk Prior Work Fine-Resolution Diagnosis Routing Algorithm Reconfiguration Algorithm Experimental Evaluation

8 Prior Art Permanent Fault Diagnosis Entire regions/routers [Puente 4] One or more bidirectional links [Fick 9] Dedicated testing and high overhead Reconfiguration around Faults Based on routing on irregular networks [Aisopos 11] Constrained by number and location of faults [Flich 7, Fukushima 9] lost nodes in a 64-node CMP ~2/3 rd loss of on-chip processor nodes # of permanent faults 8/45 8/22

9 9/45 9/22 Detection and Diagnosis Mechanism Inspired by Ghofrani et al., VTS 12 Diagnose both datapath and control faults at low area overhead (< 3%) No runtime performance overhead when NoC is fault-free Datapath faults End-to-end software based Scoreboard to collect symptoms of corrupted data host core (software score-board) router core NI core ECC router core NI core NI NI router router data packet counting and timeout error Control faults Distributed hardware based Counting and timeout techniques

10 1/45 1/22 Fine-Resolution Fault Diagnosis Packets are augmented with ECC [Shamshiri et al., ITC 11] Erroneous transmission are reported to SW supervisor node A B Differentiable from network ends A B Undistinguishable from network ends Finest granularity of: Routing reconfiguration End-to-end diagnosis

11 11/45 11/22 Fine-Grain Fault Model upstream router crossbar A B Undistinguishable: O/P port buffer downstream router crossbar Unidirectional link I/P port Crossbar contacts datapath segment packets entering KEY OBSERVATION: disable only ONE unidirectional link on most faults same packets leaving % faults localized to a single datapath segment, or a unidirectional link Error!! Error!! Error!!

12 Benefits of the Fine-Grain Fault Model Path diversity fault! X Connectivity connected! 1 2 to fault! fault! disconnected! from node node Fault manifestation Coarse-grain fault model Fine-grain fault model 12/45 12/22

13 13/45 13/22 Rest of This Talk Prior Work Fine-Resolution Diagnosis Routing Algorithm Reconfiguration Algorithm Experimental Evaluation

14 Routing in Irregular Networks Routing algorithm should disable paths that lead to deadlock Up*/down* routing disables turns to avoid deadlock 1. Construct spanning tree rooted at a node (assumes bidirectional links) 2. Mark links towards root: up (down otherwise) 3. Disable all down up turns 4. Follow up links towards root and down links to destination root darker closer to root dist dist dist 1 dist 2 root down up down up 1 towards root node up down down up away from root node down up 3 Up*/down* does not work with unidirectional links disabled turns 14/45 14/22

15 Routing with Unidirectional Links Separate spanning trees for up (up-tree) and down (down-tree) links Up-tree: unidirectional links towards root Down-tree: unidirectional links away from root Consistent ordering/labeling: no link marked both up and down As links are marked according to up*/down* principle (and no conflicts) udirec routing is deadlock free (disable down up turns) Network is connected if both trees are complete up-tree 1 to 2 {2 nd } {1 st } { th } root to conflict! down-tree from from 1 2 {1 st } {2 nd } { th } root 15/45 15/22

16 Growing Trees Concurrently Up-tree and down-tree can be constructed: Independently: may lead to inconsistent marking Concurrently: consistent labeling ensured by construction Grow tree beyond a node only if reachable by both up-tree and down-tree Choice of root node affects connectivity Step#2 Step#1 root 1 2 halt down forward down up { th } {1 st } down up {2 nd } COMPLETE disable down up turn halt down Step#1 1 2 {-} {-} down { th } up root halt up INCOMPLETE 16/45 16/22

17 17/45 17/22 Reconfiguration Overview Integrated with the software implementation of diagnosis scheme Suspension of network operation on fault detection Diagnosis of fault site via software scoreboard Supervisor core NI core NI Determination of surviving topology at supervisor Calculation of new routes in software Selection of root that maximizes connectivity Distribution of new deadlock-free routes to routers Resumption of operation from uncorrupted state router core NI router router core NI router

18 18/45 18/22 Reconfiguration Algorithm Root selection process Exhaustive search of the optimal root node Optimality of the root node Based on number of connected nodes (in our experiments) Based on critical functionality within sub-network nodes=5 nodes=3 root root root 7 8 nodes=1 root trial #1: node 1 is root root trial #2: node 2 is root root trial #6: node 6 is root root trial #8: node 6 is root local winner! local winner! local winner!

19 Reconfiguration Implementation Permanent faults are rare occurrences Routing table data = 4 (directions) * 64 (destinations) * 64 (RTs) < 2 KB Distribute using unidirectional ring of 1-control and 1-data wire unidirectional ring router RT proc proc Routing table 1-bit ctrl wire 1-bit data wire RT RT proc from previous controller RT ctrl data Serial transfer of control and data proc Distributed controller Supervisor node from proc data ctrl decode/ forward ctrl data to next controller assemble /forward update RT 19/45 19/22

20 2/45 2/22 avg. no. of dropped nodes Reliability Results 1/3 rd dropped nodes 2x less partitioning up*/down* udirec injected faults packet delivery rate up*/down* udirec injected faults As faults accumulate networks become disconnected udirec looses fewer nodes and partitions into fewer networks

21 avg. network latency Performance Results 7% lower latency path diversity dominates up*/down* udirec connectivity dominates saturation throughput up*/down* udirec injected faults injected faults Initially latency degrades gracefully; at higher fault rates up*/down* drops much more nodes, hence lower latency udirec consistently delivers higher throughput >2% higher throughput 21/45 21/22

22 22/45 22/22 Conclusions Proposed udirec: a unified diagnosis and reconfiguration solution Fine-grained fault diagnosis, a novel fault model, a deadlock-free routing algorithm and a software-based reconfiguration algorithm udirec drops only 1/3 rd nodes compared to state-of-the-art and delivers >2% higher throughput beyond 15 faults udirec incurs less than 1% wiring overhead 1/3 rd loss of nodes >2% higher

23 Thank you! Questions? 23/45 23/22