Data backup

a critical customer need

Drawbacks of existing backup solutions

• Requires user's intervention
• Same locality (accident ?)

Cloud solutions : 

• High load on servers
• One cloud provider (confidentiality ?)

External hard drive  solutions : 

Introducing peer usage

Aims at :

• No server related costs
• Spreading data over different locations

Problems : 

• Need to trust remote peers
• Peers' intermittent online behavior 

How to rely on peers for data backup ?

A peer-assisted approach

The Solution :

Main issues : 

• Data availability
• Bandwidth allocation
• Data placement

A peer-assisted approach

Data availability

a : peer availability (probability to be online)

s : number of data fragments on the server

p(s) : number of fragments stored on peers for a successful backup.

k : total number of data fragments

p(s) is optimized to guarantee a good data availability depending on a, k, s

Bandwidth allocation

A peer-assisted approach

restore crashed nodes > backup new fragments

(critical)

fragments close to completion are prioritized

Opportunistic or pessimistic strategies :
       example of data maintenance.

Bandwidth allocation

A peer-assisted approach

Example of Data Maintenance :

Successful Backup condition : x ≥ p(s)

Optimization phase : reducing servers' load

A peer-assisted approach

Successful Backup condition : x ≥ p(s)

Servers are temporary storage.

1. A fragment of the server is flagged
2. This fragment is uploaded to a peer
3. We check that x ≥ p(s-1), successful backup condition
4. If so the flagged fragment is removed from the server
5. Otherwise we keep on uploading fragments to peers

Data Placement

A peer-assisted approach

Random :
       The fragment is stored to a random peer that doesn't already have a fragment from the same peer.

Symmetric Selective :
       Peers are divided into clusters, one can only store a fragment on a peer within the same cluster and if reciprocity is respected.