HydraFS

HydraFS

• 1.
  Cristian Ungureanu, BenjaminAtkin, Akshat Aranya, Salil Gokhale, Stephen Rago, Grzegorz Całkowski, Cezary Dubnicki, and Aniruddha Bohra NEC Laboratories America Presented By : G.A.Dilruk (148209B)
• 2.
   What isHYDRAstor ?  The Research Problem  Challenges  The Design of HydraFS  Evaluation  Future Enhancements
• 3.
   HYDRAstor isa content-addressable storage (CAS) system to build storage solutions.  Data Deduplication  High Throughput  Multi Storage Nodes  Data Replication
• 4.
   Main barrieris absence of a standard API to access data in HYDRAstor.  People are lazy to change existing applications.  Applications may need to deal with unique characteristics of CAS such as block immutability and high latency.  Solution to build a standard file system on top of the HYDRAstor CAS System.
• 5.
   Blocks areimmutable, so data updates are more expensive in a CAS.  Latency of the block operations is very high.  Cache misses for metadata blocks have a significant impact on performance.  Variable block size for better Deduplication.
• 7.
   High throughputof sequential reads and writes.  Minimize the number of dependent I/O operations.  Availability guarantees of HydraFS must be no worse than standard Unix file system.  File system must efficiently support both local and remote file access.
• 8.
  Challenge Design Strategy Blocksare immutable Decouple data and meta data processing though a log buffer and batch operation in meta data. High latency of the block operations Read cache and write buffer. Cache misses for metadata blocks Fixed-size caches and use admission control to limit the number of concurrent operations. Variable block size Chunking algorithm similar to Rabin Fingerprinting
• 10.
  File Server Commit Server Transaction Log Data Blocks SuperBlocks File Operation
• 11.
   Comparison ofraw device and HydraFS file system throughput for iSCSI and Hydra.
• 12.
   Hydra andHydraFS write throughput with varying duplication ratio.
• 13.
   Multiple nodesfor File server to make failover transparent and automatic.  Integrating other algorithm like Bimodal chunking for current chucking algorithm which is similar to Rabin fingerprinting.  HydraFS is acceptable for secondary storage platform for backup appliance. Strategic way to reduce I/O latency for primary storage.
• 14.
   The HydraFile system : A first approach to a distributed file system by Benjamin Gonzalez. 24th October 2005. Computer science Department Loyola University Chicago, IL 60611, USA.  https://www.necam.com/HYDRAstor/doc.cfm?t= FAQs