Elastic Buffer Flow Control for On-Chip Networks

Abstract

This paper presents elastic buffers (EBs), an efficient flow-control scheme that uses the storage already present in pipelined channels in place of explicit input virtualchannel buffers (VCBs). With this approach, the channels themselves act as distributed FIFO buffers. Without VCBs, and hence virtual channels (VCs), deadlock prevention is achieved by duplicating physical channels. We develop a channel occupancy detector to apply universal globally adaptive load-balancing (UGAL) routing to load balance traffic in networks using EBs. Using EBs results in up to 8% (12% for low-swing channels) improvement in peak throughput per unit power compared to a VC flow-control network. These gains allow for a wider network datapath to be used to offset the removal of VCBs and increase throughput for a fixed power budget. EB networks have identical zero-load latency to VC networks operating under the same frequency. The microarchitecture of an EB router is considerably simpler than a VC router because allocators and credits are not required. For 5×5 mesh routers, this results in an 18% improvement in the cycle time.

Reliable router architecture with elastic buffer for NoC architecture

“…Elastic buffer flow control can be used to eliminate router buffers while preserving buffering in the network [5]. Elastic buffer is a pipe lined flip flops with two storage location.…”

Bubble sharing: Area and energy efficient adaptive routers using centralized buffers

“…The scheme uses elastic buffer links [9] instead of credit based flow control. This allows downstream packets to not wait on the credit round trip latency, avoiding holes among subsequent flits, even with single flit staging buffers at the input.…”

Kilo-NOC

“…Kodi et al proposed a scheme called iDEAL that augments a conventional virtual-channel architecture with in-link storage, demonstrating savings in buffer area and power [17]. An alternative proposal by Michelogiannakis et al advocates a pure elasticbuffered architecture without any virtual channels [21]. To prevent protocol deadlock in the resulting wormhole-routed NOC, the scheme requires a dedicated network for each packet class.…”

“…Figure 4 shows the proposed design in the context of a MECS network. The EB, based on the design by Michelogiannakis et al [21], uses a master-slave latch combination that can store up to two flits. We integrate an EB into each drop interface along a MECS channel and augment the baseline elastic buffer with a path from the master latch to the router input port.…”

“…To avoid deadlock in the resulting network, our approach leverages the multi-drop capability of a MECS interconnect to establish a dynamically allocated escape path for blocked packets into intermediate routers along the channel. In contrast, earlier EB schemes required multiple networks or many virtual channels for deadlock-free operation, incurring significant area and wire cost [21]. In a kilo-terminal network, the proposed single-network elastic buffer architecture requires only two virtual channels and reduces router storage requirements by 8x over a baseline MECS router without QOS support and by 12x compared to a QOS-enabled design.…”