1 Optimizing multi-processor system composition Characterization Presentation November 20 th – 2007 Performing: Isaac Yarom Supervising: Mony Orbach Annual project – Semester A (2007-1)
2 General A few teams from HS DSL are developing a multi-processors system that implements a recognition algorithm (D0417, D0317, D0136). The system is design as SOPC using: Gidel ProcstarII ® Altera FPGA, implementing NiosII ® soft core based accelerators. System will be ready at the start of
3 General - The System
4 General o SOPC design provides us flexibility in later stages of the design process. o More hardware equals greater throughput. o Main constraint is fixed limited area for hardware upon the FPGA.
5 Working environment Sampling vectors (2048B needs to be checked too) - derived from specification document. Arrive at fixed rate. NiosII ® configurations might differ in cache size and C2H Effecting throughput Effecting Area Area is limited and fixed for system
6 Project Goals Developing and implementing performance study methods. Constraint is FPGA area. This will limit number of configuration to study. Determining composition of NiosII ® cores for optimal system: Symmetric/Asymmetric system NiosII Configuration Memory defined for each NiosII Optimizing the given system using the developed methods.
7 Requirments The NiosII ® processors need to be categorized by cache size and area needed. Designed and validated NiosII ® processor (and algorithm) for basic configuration (1024B cache). Other configurations need to be setup in this project. SWITCH must support at least one basic algorithm for using asymmetric cores.
8 Project expected outcomes Performance evaluation methods and tools Might be integrated with Host software. Software model (non functional simulator) + results Recommendations for system configurations to check out in hardware Dummy-system (+ performance study) Interface between SWITCH and NiosII ® will include fields to allow system asymmetric configuration. Nice to have: Theoretical analytical recommendations for hardware systems to check
9 Project expected outcomes “ Starting point ” : Basic system Configurable NiosII ® cores Configurable SWITCH Performance evaluation methods and tools Software model + recommendations Outcomes: Performance study for 2-3 configurations Optimized system
10 Workflow (2007-1)
11 Workflow (2007-2)
12 Schedule until midterm