Download
1 / 18
180 likes | 339 Vues
Paper Presentation. TRB: Tag Replication Buffer for Enhancing the Reliability of the Cache Tag Array. Yun-Chung Yang. Shuai Wang; Jie Hu; Ziavras S.G; Dept. of Electr . & Comput . Eng., New Jersey Inst. of Technol., Newark, NJ, USA
E N D
Paper Presentation TRB: Tag Replication Buffer for Enhancing the Reliability of the Cache Tag Array Yun-Chung Yang Shuai Wang; Jie Hu; Ziavras S.G; Dept. of Electr. & Comput. Eng., New Jersey Inst. of Technol., Newark, NJ, USA VLSI (ISVLSI), 2010 IEEE Computer Society Annual Symposium on
Research Tree The architecture and application of SPM. The usage of tag Scratchpad Memory : A Design Alternative for Cache On-chip memory In Embedded Systems This paper
Abstract Protecting the on-chip cache memories against soft errorshas become an increasing challenge in designing new generation reliable microprocessors. Previous efforts have mainly focused on improving the reliability of the cache data arrays. Due to its crucial importance to the correctness of cache accesses, the tag array demands high reliability against soft errors while the data array is fully protected. Exploiting the address locality of memory accesses, we propose to duplicate most recently accessed tag entries in a small Tag Replication Buffer (TRB) thus to protect the information integrity of the tag array in the data cache with low performance, energy and area overheads. A Selective-TRBscheme is further proposed to protect only tag entries of dirty cache lines. The experimental results show that the Selective-TRB scheme achieves a higher access-with-replica(AWR) rate of 97.4% for the dirty-cache line tags. To provide a comprehensive evaluation of the tag-array reliability, we also conduct an architectural vulnerability factor (AVF) analysis for the tag array and propose a refined metric, detected-without-replica-AVF (DOR-AVF), which combines the AVF and AWR analysis. Based on our DOR-AVF analysis, a TRB scheme with early write-back (EWB) is proposed, which achieves a zero DOR-AVF at a negligible performance overhead.
Related Work Cat – caching address tags: a technique for reducing area cost of on-chip caches Computing architectural vulnerability factors for address-based structures Fault-tolerant content addressable memory On the characterization and optimization of on-chip cache reliability against soft errors Enhancing data cache reliability by the additional of a small fully-associative replication cache Fault–tolerant cam architectures: A design framework This paper
Introduction • Ionizing radiation => single-event upsets(SEUs), also known as soft errors. This could cause an incorrect data value to be read out from cache, which may crash the computation/communication. • The old-work to solve this problem • Parity code – protect the on-chip L1 caches(Intel Itanium 2, IBM Power6 processors) • Hamming code – widely adopted in L2/L3 caches • ECC(Error correct code) • SEC-DED(single error correction and double error detection)
The Proposed Method • Tag Replication Buffer(TRB) – a small buffer replicas of frequently accessed tag entries for L1 cache. • Selective-TRB – duplicate tag entries for dirty cache lines. • Vulnerability Factor Analysis(AVF) • Detected-without-replica-AVF • Early Write-back(Selective-TRB-EWB)
Architecture of TRB • Two pointer in the TRB • The set pointer => indicate the original tag entry • The way pointer => indicate the way of the original tag entry in a set-associative cache, which is not needed in a directly-mapped cache. • The copy bit indicate the tag has a replica or not. 1 is yes, 0 otherwise. • The valid bit is to indicate whether it is a valid or invalid tag replica.
TRB Mechanism • Deal Soft Error • Single-bit error, this kind of error can be detected by parity coding, but cannot be recovered. We can recover the error with TRB when the copy bit is 1. • Access-with-replica(AWR), ratio of the tag access with a replica in the TRB. • Duplicate • At the time when a cache line is brought into L2 cache. • With TRB miss. • Replacement • LRU+ or FIFO+ • Find invalidated entry instead of just doing LRU or FIFO.
Selective TRB • Improve the efficiency of protection by protecting only the dirty cache line. • Because the clean cache lines have their copies in L2 cache. And can be recoverable using ECC. • Better AWR rate compared with TRB, and also reduce the number of duplications.
AVF of Tag Array • DOR-AVF(Detected without Replica AVF) • Solution: Selective TRB with early write-back. • Duplicate only dirty cache line. • When replicate happens, its corresponding cache line will be forced to write back to L2. • 100% AWR in order to reduce the DOR-AVF to zero.
Experiment Result • SimpleScalar V3.0to model a microprocessor similar Alpha 21364. • SPEC CPU2000 as benchmark.
AWR rate in different TRB size • 8-entry TRB – 69.9% • 16-entry TRB – 82.7% • 32-entry TRB – 91.5%
Replacement Policies • LRU • AWR 91.5%. • Implement complexity higher. • FIFO • AWR 90.0% • LRU+ • AWR 92.1% • FIFO+ • AWR 91.0% • FIFO+ policy is chosen, due to the overhead of LRU+.
Three TRB policies • 33% dirty, 66% clean cache line. • 91% -> 97.4% AWR.
DOR-AVF between three methods • TRB: 31.7% -> 22.6% • Selective TRB: 31.7% -> 16.7%
Conclusion • Propose Tag Replication Buffer. • Selective TRB to improve reliability. • Selective TRB-EWB
My comment • This paper is so well organized. • The experiment is also very clear. • We can know that protecting cache information is a important topic as well.
