Handling Shared Variable Synchronization in Multi-core Network-on-Chips with Distributed Memory

• 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.

Ämnesord

TEKNIK OCH TEKNOLOGIER  -- Elektroteknik och elektronik (hsv//swe)

ENGINEERING AND TECHNOLOGY  -- Electrical Engineering, Electronic Engineering, Information Engineering (hsv//eng)

Nyckelord

900 MHz

Buffer utilization

Critical Paths

Dedicated hardware

Distributed Memory

Head of line blocking

Multi core

Network-on-chips

On-chip networks

Performance penalties

Process synchronization

Processor-memory

Remote processors

Salient features

Shared variables

Small area

Synchronization cost

Synthetic workloads

Publikations- och innehållstyp

ref (ämneskategori)

kon (ämneskategori)