Globally Synchronized Frames for guaranteed quality-of-service in on-chip networks
Journal of Parallel and Distributed Computing
Future chip multiprocessors (CMPs) may have hundreds to thousands of threads competing to access shared resources, and will require quality-of-service (QoS) support to improve system utilization. This paper introduces Globally-Synchronized Frames (GSF), a framework for providing guaranteed QoS in on-chip networks in terms of minimum bandwidth and maximum delay bound. The GSF framework can be easily integrated in a conventional virtual channel (VC) router without significantly increasing the
... ware complexity. We exploit a fast on-chip barrier network to efficiently implement GSF. Performance guarantees are verified by analysis and simulation. According to our simulations, all concurrent flows receive their guaranteed minimum share of bandwidth in compliance with a given bandwidth allocation. The average throughput degradation of GSF on an 8 × 8 mesh network is within 10% compared to the conventional best-effort VC router. (J.W. Lee). bandwidth at a memory controller is ineffective if the on-chip network does not guarantee adequate bandwidth to transport memory requests and responses. Even in a case where the onchip network is not a bandwidth bottleneck, tree saturation  can produce a tree of waiting packets that fan out from a hotspot resource, thereby penalizing remote nodes in delivering requests to the arbitration point for the hotspot resource. In this paper, we present a new scheme, Globally-Synchronized Frames (GSF), to implement QoS for multi-hop on-chip networks. GSF provides minimum bandwidth guarantees as well as bounded network delay without significantly increasing the complexity of the on-chip router. In a GSF system, time is coarsely quantized into "frames" and the system only tracks a few frames into the future to reduce time management costs. Each QoS packet from a source is tagged with a frame number indicating the desired time of future delivery to the destination. At any point of time, packets in the earliest extant frame are routed with highest priority but sources are prevented from inserting new packets into this frame. GSF exploits fast on-chip communication by using a global barrier network to determine when all packets in the earliest frame have been delivered, and then advances all sources and routers to the next frame. The next oldest frame now attains highest priority and does not admit any new packets, while resources from the previously oldest frame are recycled to form the new futuremost frame. The system can switch frames at a rate that sustains any desired set of differentiated bandwidth flows with a bounded maximum 0743-7315/$ -see front matter Jae W. Lee is an assistant professor in the Department of Semiconductor Systems Engineering at Sungkyunkwan University, Korea. His research areas include computer architecture, VLSI design, compilers, parallel programming, and computer security, and he has co-authored over a dozen papers in these areas. He led the first ASIC implementation of physical uncloneable function (PUF) at MIT and has held various engineering positions at Nvidia, Nokia, and Parakinetics. He received his M.S.