Improving coherence protocol reactiveness by trading bandwidth for latency

Menezo, Lucía G.; Puente, Valentín; Abad, Pablo; Gregorio, J.A.

doi:10.1145/2212908.2212929

Cited by 3 publications

(7 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nevertheless, multiple timeout dependent core protocol actions (such as persistent request) and semi-centralized arbitration renders this protocol unstable for memory intensive applications. 42 Rainbow is conceived as an extension to Reference 15 which was proposed assuming a single CMP coherence protocol, like many others. 30 A performance comparison with this kind of protocols seems not useful for the discussion (in the context of the class of system considered).…”

Section: Counterpart: Hyper-transport Like Coherence Protocolmentioning

confidence: 99%

Rainbow: A composable coherence protocol for multi‐chip servers

Menezo

Puente

Gregorio

2020

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

The use of multi-chip modules (MCM) and/or multi-socket boards is the most suitable approach to increase the computation density of servers while keep chip yield attained. This article introduces a new coherence protocol suitable, in terms of complexity and scalability, for this class of systems. The proposal uses two complementary ideas: (1) A mechanism that dissociates complexity from performance by means of colored-token counting, (2) A construct that optimizes performance and cost by means of two functionally symmetrical structures working in the last level cache of each chip and each memory controller. The coordinated work of both structures minimizes the coherence-related effects on the average memory latency perceived by the processor. Our proposal is able to improve on the performance of a HyperTransport-like coherence protocol by from 25% to 60%.

show abstract

Section: Counterpart: Hyper-transport Like Coherence Protocolmentioning

confidence: 99%

Rainbow: A composable coherence protocol for multi‐chip servers

Menezo

Puente

Gregorio

2020

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

show abstract

“…As a consequence, the chances of evicting an actively shared entry are higher than in numerical applications and consequently so too are the chances of requiring a multicast to reconstruct these entries. Nevertheless, even in the most adverse (and unpractical) directory configurations, this increment is below 20%, which is substantially less than in broadcast coherence protocols [10][14] [25].…”

Section: Sensitivity To Directory Size In a Realistic Private Cachementioning

confidence: 99%

“…Since the directory behaves like a serialization point, concurrent operations initiated by different processors in the same cache block will never end up suffering starvation. In this way, the directory coherence controller will avoid these problems without requiring persistent request [14] or added token tracking facilities [25] [10]. Above all, token counting is useful to filter most of the multicast messages and speed up the directory entry reconstruction.…”

Section: A Conceptual Approachmentioning

confidence: 99%

“…We will assume that the routers in the network can handle multicast traffic natively [29]. Although for this size of system a broadcast protocol might perform better [25][10], our objective is to prove that MOSAIC is capable of overcoming classic directory limitations, which will be necessary with a much higher number of cores in the CMP. Nevertheless, to evaluate systems with tens or hundreds of cores is unfeasible with current evaluation tools because of the computational effort of such a task and the limited availability of scalable workloads.…”

Section: A System Configurationmentioning

confidence: 99%

“…The first step in this analysis is to quantify how directory cost reduction influences on-chip bandwidth consumption. If the network is using routers with support for handling multicast traffic [29], the real measure of bandwidth and energy consumption for the interconnection network is given by the average link utilization and not the end-point traffic consumption [25]. Fig.…”

Section: Sensitivity To Directory Size In a Realistic Private Cachementioning

confidence: 99%

See 2 more Smart Citations

Vectorization past dependent branches through speculation

Menezo

Puente

Gregorio

2013

Proceedings of the 22nd International Conference on Parallel Architectures and Compilation Techniques

Self Cite

View full text Add to dashboard Cite

This paper introduces a new coherence protocol that addresses the challenges of complex multilevel cache hierarchies in future many-core systems. In order to keep coherence protocol complexity bounded, inclusiveness is required to track coherence information across levels in this type of systems, but this might introduce unsustainable costs for directory structures. Cost reduction decisions taken to reduce this complexity may introduce artificial inefficiencies in the on-chip cache hierarchy, especially when the number of cores and private caches size is large. The coherence protocol presented in this work, denoted MOSAIC, introduces a new approach to tackle this problem. In energy terms, the protocol scales like a conventional directory coherence protocol, but relaxes the shared information inclusiveness. This allows the performance implications of directory size and associativity reduction to be overcome. Contrary to the common belief that inclusiveness is inescapable when attempting to maintain complexity constrained, MOSAIC is even simpler than a conventional directory. The results of our evaluation show that the approach is quite insensitive, in terms of performance and energy expenditure, to the size and associativity of the directory.

show abstract