Exploiting Properties of CMP Cache Traffic in Designing Hybrid Packet/Circuit Switched NoCsAbousamra, Ahmed (2013) Exploiting Properties of CMP Cache Traffic in Designing Hybrid Packet/Circuit Switched NoCs. Doctoral Dissertation, University of Pittsburgh. (Unpublished)
AbstractChip multiprocessors with few to tens of processing cores are already commercially available. Increased scaling of technology is making it feasible to integrate even more cores on a single chip. Providing the cores with fast access to data is vital to overall system performance. When a core requires access to a piece of data, the core's private cache memory is searched first. If a miss occurs, the data is looked up in the next level(s) of the memory hierarchy, where often one or more levels of cache are shared between two or more cores. Communication between the cores and the slices of the on-chip shared cache is carried through the network-on-chip(NoC). Interestingly, the cache and NoC mutually affect the operation of each other; communication over the NoC affects the access latency of cache data, while the cache organization generates the coherence and data messages, thus affecting the communication patterns and latency over the NoC. This thesis considers hybrid packet/circuit switched NoCs, i.e., packet switched NoCs enhanced with the ability to configure circuits. The communication and performance benefit that come from using circuits is predicated on amortizing the time cost incurred for configuring the circuits. To address this challenge, NoC designs are proposed that take advantage of properties of the cache traffic, namely temporal locality and predictability, to amortize or hide the circuit configuration time cost. First, a coarse-grained circuit configuration policy is proposed that exploits the temporal locality in the cache traffic to periodically configure circuits for the heavily communicating nodes. This allows the design of a locality-aware cache that promotes temporal communication locality through data placement, while designing suitable data replacement and migration policies. Next, a fine-grained configuration policy, called Déjà Vu switching, is proposed for leveraging predictability of data messages by initiating a circuit configuration as soon as a cache hit is detected and before the data becomes available. Its benefit is demonstrated for saving interconnect energy in multi-plane NoCs. Finally, a more proactive configuration policy is proposed for fast caches, where circuit reservations are initiated by request messages, which can greatly improve communication latency and system performance. Share
Details
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[feed]](/images/feed-icon-32x32.png){kind=icon}