RACS: A Case for Cloud Storage Diversity
1
Hussam Abu-Libdeh,Lonnie PrincehouseHakim WeatherspoonCornell University
Problem
Vendor Lock-in:Longer time they use one cloud provider more data stored in this cloud provider harder and more expensive to transfer their data out (transfer in $$$$, transfer out $$$$) Lock in one cloud provider More vulnerable for price hikes/data center failures(Two main problems)
Internet Archive--Data Transfer
Internet Archive—Read/Write
Stuck in one cloud provider Outage and operation failure:
physical failure Google Cloud unavailable Microsoft data center failure in Oct. 2009
Economic failures: budget now allowed Emerging new cheaper cloud Current cloud provider increase price
How to guard against Vendor Lock-in First thought:
Why not replicate their data into multiple providers? ▪ Could not; three replicas, too expensive
Another way to create redundancy: ▪ Error Correcting code (such as RAID 5)
Preliminary knowledge: RAID 5
Redundant Array of Cloud Storage Consider each cloud provider as a disk
in RAID.
Design
Implemented as Proxy between client application and n- cloud storages Strip data into m
pieces, put them into m clouds, and generate (n-m) redundant data.
Distributed RACS
Failures Recovery
Error Coding Recovery
Cost of hosting on the cloud
Cost of Switching Vendors
Tolerant a Vendor Price Hike
Different Scenarios (1)
Upload Snapshot
Different Scenarios
Vendor Migration
Different Scenarios (3)
Restore snapshot
RACS performance
Conclusion
Identified an important problem: vendor lock-in
Proposed Redundant Array of Cloud Storage to solve this problem using erasure coding.
Tradeoff between overhead expense and vendor mobility
Simulations and experiments to prove the virtue for RACS.
Reference
[1] RACS: A Case for Cloud Storage Diversity. Hussam Abu-Libdeh, Lonnie Princehouse, Hakim Weatherspoon. ACM Symposium on Cloud Computing (SOCC). June 2010, Indianapolis, IN.