Token tenure: PATCHing token counting using directory-based cache coherence

Abstract

Follow this and additional works at: http://repository.upenn.edu/cis_reports Token tenure: PATCHing token counting using directory-based cache coherence Raghavan, A.; Blundell, C.; Martin, M.M.K. Copyright 2008 IEEE. Reprinted from MICRO-41. 41st IEEE/ACM International Symposium on Microarchitecture, 8-12 Nov. 2008 ,pages 47-58. This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. Traditional coherence protocols present a set of difficult tradeoffs: the reliance of snoopy protocols on broadcast and ordered interconnects limits their scalability, while directory protocols incur a performance penalty on sharing misses due to indirection. This work introduces PATCH (Predictive/Adaptive Token Counting Hybrid), a coherence protocol that provides the scalability of directory protocols while opportunistically sending direct requests to reduce sharing latency. PATCH extends a standard directory protocol to track tokens and use token counting rules for enforcing coherence permissions. Token counting allows PATCH to support direct requests on an unordered interconnect, while a mechanism called token tenure uses local processor timeouts and the directorypsilas per-block point of ordering at the home node to guarantee forward progress without relying on broadcast. PATCH makes three main contributions. First, PATCH introduces token tenure, which provides broadcast-free forward progress for token counting protocols. Second, PATCH deprioritizes best-effort direct requests to match or exceed the performance of directory protocols without restricting scalability. Finally, PATCH provides greater scalability than directory protocols when using inexact encodings of sharers because only processors holding tokens need to acknowledge requests. Overall, PATCH is a ldquoone-size-fits-allrdquo coherence protocol that dynamically adapts to work well for small systems, large systems, and anywhere in between. CommentsToken tenure: PATCHing token counting using directory-based cache coherence Raghavan, A.; Blundell, C. This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org. By choosing to view this d...

Analyzing Cache Coherence Protocols for Server Consolidation

“…Token-based and directory-based coherence mechanisms have already been successfully combined in a single pro-tocol [6]. Therefore, it is reasonable to wonder whether V H B , which uses token counting in addition to the directory information, is also a token-based protocol with a correctness substrate that would prevent any deadlock or starvation situations.…”

Towards the ideal on-chip fabric for 1-to-many and many-to-1 communication

“…However, the required storage increases area and energy costs as core counts scale. More scalable directory protocols [1,9,12,26,33], including commercial designs like Intel's Quick Path Interconnect (QPI) protocols [1] and AMD's HyperTransport TM Assist [12] incorporate partial directories to consume less storage than a full-bit directory, and rely on a combination of broadcasts, multicasts, and direct requests to maintain coherence.…”

“…1-to-M communication occurs in broadcasts and multicast requests [1,6,11,29,36]. M-to-1 communication occurs in acknowledgements [1,11] or token collection [29,33] in protocols to maintain ordering. In the on-chip domain, conventional wisdom seems to dictate that 1-to-M and M-to-1 traffic should be avoided, assuming the on-chip network will not be able to handle such high bandwidth.…”

“…M-to-1 communication flow occurs in shared memory coherence protocols, in the form of acknowledgements [1,11] or tokens [29,33], and in message passing domains, such as barrier synchronization [4,10,18,39]. There has been no on-chip solution to tackle the former, to the best of our knowledge.…”

Switch-based packing technique to reduce traffic and latency in token coherence

“…When this situation happens, Token Coherence must use a starvation prevention mechanism that is able to guarantee the resolution of all cache misses. To date, different proposals of starvation prevention mechanisms have been made, but they present serious drawbacks: persistent requests [8] are broadcast-based and inefficient; token tenure [9] is not based on broadcast, but like persistent requests it is quite inefficient; priority requests [10] efficiently manage tokens, but they are broadcast-based. Thus, although one of the mechanisms (priority requests) is efficient, it is based on broadcast.…”