low-overheadbyzantinefault-tolerantstorage-資料下載頁
2025-07-17 15:32本頁面
【正文】 ricks, 22 1 2 3 4 5 6 7 8 9 1001020304050607080Write throughput Number of faults tolerated (f) Bandwidth (MB/s) James Hendricks, 23 1 2 3 4 5 6 7 8 9 1001020304050607080Write throughput Number of faults tolerated (f) Bandwidth (MB/s) 64 kB f +1 16 kB fragments James Hendricks, 24 1 2 3 4 5 6 7 8 9 100123456767Write response time Latency (ms) Number of faults tolerated (f ) A B C D RPC Hash Encode Fingerprint Overhead breakdown (f =10) James Hendricks, 25 1 2 3 4 5 6 7 8 9 10020406080100120Read throughput Bandwidth (MB/s) Number of faults tolerated (f) Crashonly replication based Our protocol Crashonly erasurecoded James Hendricks, 26 1 2 3 4 5 6 7 8 9 1000 . 511 . 522 . 533 . 544 . 55Read response time Latency (ms) Number of faults tolerated (f ) D. Replicated B. Our protocol A. Erasurecoded A B C D RPC Hash Encode Fingerprint Overhead breakdown (f =10) James Hendricks, 27 Conclusions Byzantine faulttolerant storage can rival crashonly storage performance We present a low overhead Byzantine faulttolerant erasurecoded block storage protocol and prototype Write overhead: 2round, hash and fingerprint Read overhead: hash and fingerprint Close to performance of systems that tolerate only crashes for reads and large writes
物理相關(guān)推薦
鄂ICP備17016276號-1