Adaptive selection of parity chunk update methods in RAID-enabled SSDs - Parallel and Distributed Systems group Accéder directement au contenu
Communication Dans Un Congrès Année : 2024

Adaptive selection of parity chunk update methods in RAID-enabled SSDs

Résumé

RAID-enabled SSDs commonly have unbalanced I/O workloads on their components (e.g., on their SSD channels), and updating parity requires issuing a number of additional read requests (termed as pre-reads), which causes long tail latency. To address this issue, we introduce an adaptive scheme for selecting the update method on parity chunks. Our scheme uses the most suitable routine for updating parity to better balance I/O accesses over all channels of RAID-enabled SSDs. In particular, it can adaptively select routines including read-modify-write (RMW) and read-construct-write (RCW), when renewing data stripes. To this end, we build a mathematical model to assess the time required for updating parity chunks with different update routines by considering the number of pre-read requests and the blocked I/O traffic on the affected SSD channels. As a result, it can determine whether the update of parity chunk should be performed with RMW or RCW during I/O scheduling. Trace-driven experiments illustrate that the proposed scheme can balance the workloads across channels, as well as reduce the long-tail latency of I/O requests by up to 24.8% at the 99.9th percentile and the overall I/O time by 13.5% on average, in contrast to existing approaches.
Fichier principal
Vignette du fichier
Adaptive_Selection_of_Parity_Chunk_Update_Methods_in_RAID-enabled_SSDs.pdf (1.47 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04569056 , version 1 (06-05-2024)

Identifiants

  • HAL Id : hal-04569056 , version 1

Citer

Fan Yang, Zhigang Cai, Jun Li, Balazs Gerofi, François Trahay, et al.. Adaptive selection of parity chunk update methods in RAID-enabled SSDs. MSST 2024 - The 38th International Conference on Massive Storage Systems and Technology, Jun 2024, Santa Clara, United States. ⟨hal-04569056⟩
0 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More