Automated P2P Backup

1. Automated P2P Backup Group 1Anderson, Bowers, Johnson, Walker

2. Motivation • Hard drive space is cheap • Network connectivity is cheap • Losing data is expensive • We’d like to pool our resources and easily collectively maintain backups

3. Background – Distributed Hashes • Most academic P2P systems are built on “distributed hash tables” • Ask “the system” for a key, and get the content back • How the Distributed Hash and the hash lookup is implemented characterizes the P2P system

4. Related Work – P2P • Pastry • Each node has an ID, messages routed to node nearest the hash key • N-order graph used to route • OceanStore • Write-once, has versioning • Emphasizes local storage

5. Related Work – P2P (cont.) • CHORD • Routes similarly to Pastry • Circular routing space • Freenet • Write-once • Many security and anonymity features • All resource encrypted by their hash key

6. Related Work – Distributed Backup • Distributed Internet Backup System • Not P2P – uses direct connections • Encrypts user data on other drives • Pastiche • Built on Pastry • PAST • Also built on Pastry • pStore • Uses own P2P architecture

7. Design

8. P2P Adapter • Abstracts interfacing with a P2P system’s distributed hash • Writing another adapter could make this work with another system • We used Freenet because it is the only working P2P system publicly available • When backing up a file, returns a key that the adapter can later use to retrieve the file • P2P system specific • Freenet example: SSK@iuhgwe78gfq93f73f:/something/path/file.txt • Done in Python

9. Engine • Implements backup and retrieval logic • Drives the P2P adapter to insert and retrieve files • Stores keys from P2P adapter for retrieval • Done in Java

10. UI • Allows the user to select files or directories for backup % mark_dir > backup > done • Allows user to initiate backup or retrieval based on selection • Stores selections in “backup.txt” as a comma-delimited file storing the filename, date of last backup, and retrieval key (if any) • Done in Java

11. Example Run bowersj2@arctic> ls –l total 8 drwx------ 2 bowersj2 student 4096 Apr 16 13:58 dir -rw------- 1 bowersj2 student 27 Apr 16 13:58 example bowersj2@arctic> java Backup_UI % mark_dir > dir > done % mark_file > example > done % backup Backing up dir/one... retrieval key freenet:SSK@iuhgwe78gfq93f73f/3 Backing up dir/two... retrieval key freenet:SSK@iuhgwe78gfq93f73f/4 Backing up example... retrieval key freenet:SSK@iuhgwe78gfq93f73f/5 % exit

12. Results • One limit of the Freenet library used was that files must be no larger then 64KB • Not fundamental to Freenet • A Freenet file takes approximately 4-5 seconds to insert into the system • Retrieval was very fast since it was always from the local drive cache

13. Design Benefits • Careful design allows each component to be implemented in any language • UI and Engine communicate through backup.txt • Engine and P2P adapter communicate through command lines • Problems getting other P2P systems running • Most not publicly available yet • PASTRY could not be compiled • Shipped source had Java exception handling errors

14. Conclusion • Modern P2P systems will provide a good substrate for this sort of application • When they are released and working! • Writing a basic version of this kind of application is fairly easy • Effectiveness depends on the underlying P2P system • Freenet doesn’t chunk files, some P2P systems do • Freenet has no retention guarantees, some P2P systems do • Freenet natively prevents snooping by other users, some P2P systems don’t