Built‐in fast gather control network for efficient support of coherence protocols

Abstract

Future chip multiprocessors will include hundreds of cores organised in a tile-based design pattern. These systems commonly employ a shared memory programming model, thus needing a coherence protocol to keep data consistent on the various levels of the cache hierarchy. Usually an invalidation-based protocol is used, where shared copies are invalidated before a write operation. In this study, the authors propose a NoC re-organisation in which a small and fast dedicated control network is used to transmit acknowledgement messages related to the invalidation process, thus relieving the NoC from a considerable percentage of traffic. The dedicated control network is evaluated both with full map directories and with a broadcast-based protocol (Hammer). Experimental evaluation shows significant gains in performance. With a low area overhead (<2.5%), the control network reduces NoC traffic and miss latency, thus reducing execution time up to 16%. Simulation results show a reduction of network traffic up to 80% and a reduction of store and load miss latency up to 70 and 40%, respectively.

Runtime home mapping for effective memory resource usage

“…Once all L2 banks trigger their output signals, the output of the AND tree will notify the L2-0 and thus will act as a global ACK message. Two different implementations of the GCN were used in our previous works to speedup the ACKs sent by L1 caches to the L1 requestor in directory-based and broadcast-based coherence protocols [3,5,4]. In this work, we provide a detailed description of the GCN logic and extend its use to collect acknowledgements generated by and destined to different levels of the cache hierarchy and to deliver simple control information together with the global ACK.…”

A Lightweight Network of IDs to Quickly Deliver Simple Control Messages

“…To address these issues, we implemented the logical behavior of the combinational implementation with sequential logic [4]. This second version of the Gather Network uses sequential modules at each switch, which communicate transmitting the ID of the acknowledgement's destination node, thus improving the scalability since the wiring requirements are reduced to the logarithm in base two of the number of tiles in the CMP.…”

“…However, these messages are typically transmitted through the NoC using the same short message structure of requests, thus wasting NoC resources and power to transmit a packet in which most of the fields are not used or redundant. Basing on these considerations, we recently proposed a lightweight control network, called the Gather Network [4], used to collect many-to-one acknowledgements in invalidation-based coherence protocols. In this paper we extend the Gather Network and adapt it to transmit unicast acknowledgements (e.g.…”

Towards Efficient Dynamic LLC Home Bank Mapping with NoC-Level Support

“…The GCN we assume in this work has the same logic behavior of the AND trees but is implemented with sequential logic, which reduces wiring requirements. An extensive discussion on the sequential GCN is published in [4].…”