11. |
- Candaele, Bernard, et al.
(author)
-
The MOSART Mapping Optimization for multi-core Architectures
- 2011
-
In: VLSI 2010 Annual Symposium. - Dordrecht : Springer Publishing Company. ; , s. 181-195
-
Conference paper (peer-reviewed)abstract
- MOSART project addresses two main challenges of prevailing architectures: (i) Theglobal interconnect and memory bottleneck due to a single, globally shared memorywith high access times and power consumption; (ii) The difficulties in programmingheterogeneous, multi-core platforms MOSART aims to overcome these through amulti-core architecture with distributed memory organization, a Network-on-Chip(NoC) communication backbone and configurable processing cores that are scaled,optimized and customized together to achieve diverse energy, performance, cost andsize requirements of different classes of applications. MOSART achieves this by:(i) Providing platform support for management of abstract data structures includingmiddleware services and a run-time data manager for NoC based communicationinfrastructure; (ii) Developing tool support for parallelizing and mapping applicationson the multi-core target platform and customizing the processing cores for theapplication.
|
|
12. |
- Chen, Xiaowen, et al.
(author)
-
Area and Performance Optimization of Barrier Synchronization on Multi-core Network-on-Chips
- 2010
-
In: 3rd IEEE International Conference on Computer and Electrical Engineering (ICCEE).
-
Conference paper (peer-reviewed)abstract
- Barrier synchronization is commonly and widelyused to synchronize the execution of parallel processor coreson multi-core Network-on-Chips (NoCs). Since its globalnature may cause heavy serialization resulting in largeperformance penalty, barrier synchronization should becarefully designed to have low latency communication and tominimize overall completion time. Therefore, in the paper, wepropose a fast barrier synchronization mechanism, targetingMulti-core NoCs. The fast barrier synchronization mechanismincludes a dedicated hardware module, named Fast BarrierSynchronizer (FBS), integrated with each processor node. Itoffers a set of barrier counters and can concurrently processsynchronization requests issued by the local node and remotenodes via the on-chip network. The salient feature of our fastbarrier synchronization mechanism is that, once the barriercondition is reached, the “barrier release” acknowledgement isrouted to all processor nodes in a broadcast way in order tosave chip area by avoiding storing source node informationand to minimize completion time by avoiding serialization ofbarrier releasing. Synthesis results suggest that the FBS canrun over 1 GHz in SMIC® 130nm technology with small areaoverhead. We implemented a FBS-enhanced multi-core NoCarchitecture on our FPGA platform using the Xilinx® Virtex 5as the FPGA chip. FPGA utilization and simulation resultsshow that our fast barrier synchronization demonstrates botharea and performance advantages over the barriersynchronization counterpart with unicast barrier releasing.
|
|
13. |
- Chen, Xiaowen, et al.
(author)
-
Handling Shared Variable Synchronization in Multi-core Network-on-Chips with Distributed Memory
- 2010
-
In: Proceedings. - 9781424466832 ; , s. 467-472
-
Conference paper (peer-reviewed)abstract
- Parallelized shared variable applications running on multi-core Network-on-Chips(NoCs) require efficient support for synchronization, since communication is on the critical path of system performance and contended synchronization requests may cause large performance penalty. In this paper, we propose a dedicated hardware module forsynchronization management. This module is called Synchronization Handler (SH), integrated with each processor-memory node on the multi-core NoCs. It uses two physical buffers to concurrently process synchronization requests issued by the local processor and remote processors via the on-chip network. One salient feature is that the two physical buffers are dynamically allocated to form multiple virtual buffers (a virtual buffer is related to a shared synchronization variable) so as to improve the buffer utilization and alleviate the head-of-line blocking. Synthesis results suggest that the SH can run over 900 MHz in 130nm technology with small area overhead. To justify the SH-enhanced multicore NoCs, we employ synthetic workloads to evaluate synchronizationcost and buffer utilization, and run synchronization-intensive applications to investigate speedup. The results show that our approach is viable.
|
|
14. |
- Chen, Xiaowen, et al.
(author)
-
Multi-FPGA Implementation of a Network-on-Chip Based Many-core Architecture with Fast Barrier Synchronization Mechanism
- 2010
-
In: Proceedings of the IEEE Norchip Conference. - 9781424489732
-
Conference paper (peer-reviewed)abstract
- In this paper, we propose a fast barrier synchronization mechanism, targetingNetwork-on-Chip based manycore architectures. Its salient feature is that, once thebarrier condition is reached, the "barrier release" acknowledgement is routed to all processor nodes in a broadcast way in order to save area by avoiding storing source node information and to minimize completion time by eliminating serialization of barrierreleasing. Then, we construct a multi-FPGA platform using Xilinx® Virtex 5 as FPGA chipsand implement a NoC based many-core architecture on it. FPGA utilization and simulation results show that our mechanism demonstrates both area and performance advantages over the barrier synchronization counterpart with unicast barrier releasing.
|
|
15. |
- Chen, Xiaowen, 1982-, et al.
(author)
-
Run-time Partitioning of Hybrid Distributed Shared Memory on Multi-core Network-on-Chips
- 2010
-
In: The 3rd IEEE International Symposium on Parallel Architectures, Algorithms and Programming (PAAP 2010). - : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 39-46
-
Conference paper (peer-reviewed)abstract
- On multi-core Network-on-Chips (NoCs), mem- ories are preferably distributed and supporting Distributed Shared Memory (DSM) is essential for the sake of reusing huge amount of legacy code and easy programming. However, the DSM organization imports the inherent overhead of translating virtual memory addresses into physical memoryaddresses, resulting in negative performance. We observe that, in parallel applications, different data have different properties (private or shared). For the private data accesses, it's unnecessary to perform Virtual-to-Physical address translations. Even for the same datum, its property may be changeable in different phases of the program execution. Therefore, this paper focuses on decreasing the overhead of Virtual-to-Physical address translation and hence improving the system performance by introducing hybrid DSM organization and supporting run-time partitioning according to the data property. Thehybrid DSM organization aims at supporting fast and physical memory accesses for private data and maintaining a global and single virtual memory space for shared data. Based on the data property of parallel applications, the run-time partitioning supports changing the hybrid DSM organization during the program execution. It ensures fast physical memory addressing on private data and conventional virtual memory addressingon shared data, improving the performance of the entire system by reducing virtual-to-physical address translation overhead as much as possible. We formulate the run-timepartitioning of hybrid DSM organization in order to analyze its perfor- mance. A real DSM based multi-core NoC platform is also constructed. The experimental results of real applications show that the hybrid DSM organization with run-time partitioningdemonstrates performance advantage over the conventional DSM counterpart. The percentage of performance improve- ment depends on problem size, way of datapartitioning and computation/ communication ratio of parallel applications, network size of the system, etc. In our experiments, the maximal improvement is 34.42%, the minimal improvement 3.68%.
|
|
16. |
- Chen, Xiaowen, et al.
(author)
-
Speedup Analysis of Data-parallel Applications on Multi-core NoCs
- 2009
-
In: Proceedings of the IEEE International Conference on ASIC (ASICON). - 9781424438686 ; , s. 105-108
-
Conference paper (peer-reviewed)abstract
- As more computing cores are integrated onto a single chip, the effect of network communication latency is becoming more and more significant on Multi-core Network-onChips (NoCs). For data-parallel applications, we study the model ofparallel speedup by including network communication latency in Amdahl's law. The speedup analysis considers the effect of network topology, network size, traffic model and computation/communication ratio. We also study the speedup efficiency. In our Multi-core NoC platform, a real data-parallel application, i.e. matrix multiplication, is used to validate the analysis. Our theoretical analysis and the application results show that the speedup improvement is nonlinear and the speedup efficiency decreases as the system size is scaled up. Such analysis can be used to guide architects and programmers to improve parallel processing efficiency by reducing network latency with optimized network design and increasing computation proportion in the program.
|
|
17. |
- Chen, Xiaowen, et al.
(author)
-
Supporting Distributed Shared Memory on Multi-core Network-on-Chips Using a Dual Microcoded Controller
- 2010
-
In: Proceedings of the conference for Design Automation and Test in Europe. ; , s. 39-44
-
Conference paper (peer-reviewed)abstract
- Supporting Distributed Shared Memory (DSM) is essential for multi-coreNetwork-on-Chips for the sake of reusing huge amount of legacy code and easy programmability. We propose a microcoded controller as a hardware module in each node to connect the core, the local memory and the network. The controller is programmable where the DSM functions such as virtual-to-physical address translation,memory access and synchronization etc. are realized using microcode. To enable concurrent processing of memory requests from the local and remote cores, ourcontroller features two mini-processors, one dealing with requests from the local coreand the other from remote cores. Synthesis results suggest that the controller consumes 51k gates for the logic and can run up to 455 MHz in 130 nm technology. To evaluate its performance, we use synthetic and application workloads. Results show that, when the system size is scaled up, the delay overhead incurred by the controller may become less significant when compared with the network delay. In this way, the delay efficiency of our DSM solution is close to hardware solutions on average but still have all the flexibility of software solutions.
|
|
18. |
- Chen, Xiaowen, 1982-, et al.
(author)
-
Supporting Efficient Synchronization in Multi-core NoCs Using Dynamic Buffer Allocation Technique
- 2010
-
In: Proceedings of the IEEE Annual Symposium on VLSI. - : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 462-463
-
Conference paper (peer-reviewed)abstract
- This paper explores a dynamic buffer allocation technique to guide a distributedsynchronization architecture to support efficient synchronization on multi-core Network-on-Chips (NoCs). The synchronization architecture features two physical buffers to be able to concurrently queue and handle synchronization requests issued by the local processor and remote processors via the on-chip network. Using the dynamic bufferallocation technique, the two physical buffers are dynamically allocated to form multiple virtual buffers in order to improve buffers' utilization. Experiments are carried on to evaluate buffers' utilization.
|
|
19. |
- Chen, Zhipeng, et al.
(author)
-
A Worst Case Performance Model for TDM Virtual Circuit in NoCs
- 2010
-
In: NETWORK AND PARALLEL COMPUTING. - Berlin : Springer Berlin/Heidelberg. - 9783642156717 ; , s. 452-461
-
Conference paper (peer-reviewed)abstract
- In Network-on-Chip (NoC), Time-Division-Mutiplexing (TDM) Virtual Circuit (VC) is well recognized as being capable to provide guaranteed services in both latency and bandwidth. We propose a method of modeling TDM based VC by using Network Calculus. We derive a tight upper bound of end-to-end delay and buffer requirement for individual VC. The performance analysis using Latency-Rate server is also presented in comparsion with our Performance model for TDM Virtual Circuit in NoCs (Pemvin). We conducted experiments on comparing Pemvin to the Latency-Rate server model. Our experiment results show the improvement of Pemvin on tightening the upper bound of end-to-end delay and buffer requirement.
|
|
20. |
|
|