Fairness in Reliability: Equitable Use of Reliability Enhancement in Caches
IEEE Transactions on Emerging Topics in Computing, cilt.14, sa.3, ss.1122-1134, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14 Sayı: 3
- Basım Tarihi: 2026
- Doi Numarası: 10.1109/tetc.2026.3723060
- Dergi Adı: IEEE Transactions on Emerging Topics in Computing
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC
- Sayfa Sayıları: ss.1122-1134
- Anahtar Kelimeler: cache reallocation, dynamic fault tolerance, fairness, fault injection, Reliability, selective protection
- Marmara Üniversitesi Adresli: Evet
Özet
Continuous technology scaling has made moderncomputer systems increasingly vulnerable to soft errors, makingfault tolerance approaches increasingly important. Although suchapproaches generally provide substantial reliability enhancements, their performance and energy overheads can be prohibitive. Selective fault tolerance methods address this problemby providing sufficient reliability instead of conservative (worstcase) reliability. However, when multiple applications use thesame architecture with limited reliability enhancement capability,allocating this capability across competing applications in afair manner remains an open research problem. Our paperaddresses fairness problem in the reliability domain by focusing on cache memory. The proposed framework dynamicallyallocates ECC-protected cache blocks to concurrently runningapplications based on their error rates. To achieve this, errors aremonitored at three levels-physical, data access, and applicationeach employing a distinct reliability metric. The effectivenessof the framework is evaluated in various scenarios as partof our empirical study. Our framework successfully enhancesfairness in reliability, particularly when applications exhibitcomparable error resilience. In heterogeneous workloads, fairnessin reliability (F iR), calculated as the error rate difference acrossapplications, is improved from 0.48 to 0.29; on the other hand, forhomogeneous workloads, the F iR value is improved from 0.67 to0.30. Experimental results indicate that error rates obtained fromlower levels allow us to adjust the error rates at the applicationlevel, resulting in fairer usage of protected hardware resources.