CDStore: Toward Reliable, Secure, and Cost-Efficient Cloud Storage via
Convergent Dispersal
Download
ATC, 2015
Summary
This paper presents CDStore, a client-server based multi-cloud storage solution with reliability, security and cost-efficiency guarantees. CDStore adopts two-stage deduplication which builds on Convergence Dispersal by using deterministic content-derived hashes as inputs to secret sharing, and it can achieve both bandwidth and storage savings and be robust against side-channel attacks.
Details
- Convergence Dispersal, AONT-RS, CAONT-RS
- OAEP-based AONT to improve performance on speed
- Replace random inputs with deterministic hashes to allow deduplication
- Due to randomness, secret-sharing prohibits deduplication
- CD Replace random input with deterministic hash, thus allows deduplication
- Stronger hash key can be applied to mitigate brute-force attacks
- Exps shows that CAONT-RS generates despersed data faster than AONT-RS based instantiation
- Two-stage deduplication
- inter-users, intra-users dedup
- dedup greately reduces storage overhead in some cases
- (n, k, r) settings, (n, k) determines fault torlerance degree, (k, r) determines confidentiality degree
- can avoid side-channel attacks by making dedup patterns independent accross users’ uploads. At least (n - r) uncompromized
- Fault Torlerance on Client and Server
- offloading metadata management to server side
- Implementation
- Metadata offloading (file metadata, share metadata are distributed to all servers)
- Index management
- Multi-thread Optimization in secret encoding/decoding
- Microbenchmarks for reference
Strength
- Two-stage deduplication
- Achieve both bandwidth and storage savings
- Robust against side-channel attacks
- Convergence Dispersal + AONT + Reed-Solomon code ensures security
Weakness
-
Strong attack models (Byzantine faults) are not considered
-
Assume protected Client-Server connection (attacks like eavesdropping doesn’t work)
-
Encoding in client side (requires computation effort)
-
GC and compression are not considered
-
Load Balance (multi-VM (server) per cloud)
-
Consistency