Zbackup: open-source, encrypted, de-duplicated, compressed backups | Hacker News Reader

Zbackup: open-source, encrypted, de-duplicated, compressed backups

github.com

76 pointsfrenkel21 comments

I'm a little concerned about the use of deduplication and encryption in the same system. If done improperly, deduplication followed by encryption potentially leaks information--a passive adversary can make some good guesses at what's going in with the plaintext across backups, based on how big or small the size delta is. Maybe I'm being overly paranoid, but I didn't find a satisfactory answer in the documentation at how to plug this side-channel. Would an experienced user or the author be willing to comment?

The active version of this attack is totally worrying (deduplication between multiple user accounts, alice and eve).

But the passive version above requires a privileged network position as well as an unrealistically good idea about what was in the original backup. I agree it's a weakness, but i'm not sure it's a cause for concern

You might equally say something like "the amount of time taken to backup is a timing side-channel into the size of the deduplication cache", it's true but i don't think it immediately lead to any practical attacks.

Can you elaborate on a scenario?

ZPAQ has much to give, a good recommendation. But I also found obnam awesome for de-duplicated backups as it allows cool snapshots which can be mounted via FUSE to be browsed in Nautilus or extracted via commandline tools.

Package: obnam Version: 1.8-1 Maintainer: Lars Wirzenius <liw@liw.fi> Depends: libc6 (>= 2.6), python (>= 2.7), python (<< 2.8), python-larch (>= 1.20131130~), python-ttystatus (>= 0.23~), python-paramiko, python-tracing (>= 0.8~), python-cliapp (>= 1.20130808~), python-fuse Description-en: online and disk-based backup application Obnam makes backups. Backups can be stored on local hard disks, or online via the SSH SFTP protocol. The backup server, if used, does not require any special software, on top of SSH. . * Snapshot backups. Every generation looks like a complete snapshot, so you don't need to care about full versus incremental backups, or rotate real or virtual tapes. * Data de-duplication, across files, and backup generations. If the backup repository already contains a particular chunk of data, it will be re-used, even if it was in another file in an older backup generation. This way, you don't need to worry about moving around large files, or modifying them. * Encrypted backups, using GnuPG. * Push or pull operation, depending on what you need. You can run Obnam on the client, and push backups to the server, or on the server, and pull from the client over SFTP.

Homepage: http://liw.fi/obnam/

Ive recently found http://mattmahoney.net/dc/zpaq.html which (on paper) is as good as zbackup/bup/rdiff-backup. Haven't had time to try it - anyone her has experience?

I use it for some backups (not automatic ones, just when I have some old stuff and I want to put it "in the corner"). Very good compression ratio (beats bz2 and lzma in my cases - big parts are db dumps), but when achieving it, it's also very slow.

Be careful when feeding whole directories to it since it doesn't care about files attributes.

This looks very similar to attic, anyone able and willing to compare them for pros and cons?

I would throw bup[1] into the competition too. One pro over zbackup is the easier retrieval of single files. Bup uses the same rolling checksum method for dedupliation inspired by rsync but it uses git pack files to store the data. Some git functionalities are reimplemented to work with large number of files and large files.

[1] https://github.com/bup/bup

edit: One huge con is that you can't prune old backup data. It's really early in development but it looks promising.

It so happens that I found zbackup in the list of apt packages just yesterday. It doesn't have network support like attic, so you have to transport the entire repository of data to the backup server in order to make a backup.

In practice that's impractical for anything big. You can use rsync to do the transfer to a local clone of the data on the backup server but then the whole thing still needs to be ingested each time.

It has a rolling checksum splitter just like attic, which is unbelievably effective way to split files into chunks for deduplication. It'll work really well for database dumps, which is something that fixed-size chunking fails at miserably.

A big advantage over bup is that you can remove old backups.

Other backup software worth looking at is rdiff-backup, duplicity, burp, and obnam, all of which are in apt.

You can easily use netcat to pipe tar over the network directly to zbackup. If you want encryption over the network it's even easier:

tar -c stuff/ | ssh user@example.com zbackup backup /my/backup/repo/backups/backup-`date '+%Y-%m-%d'`

I should have been more clear. The problem is that zbackup won't work for large data sets. For example, I used to use rdiff-backup for a samba filesystem at work that had a few terabytes of files on it. The problem was that whenever someone would rename a directory all of the data in its subdirectories would have to be backed up again.

Backup software that deduplicates solves this problem wonderfully, since it doesn't make a difference if the files have been moved or not.

The problem with your approach is that I can't send several terabytes of data offsite every night. I could use the rsync trick I mentioned, but now I've got to store two copies of the data on the backup server.

Wrong.

You could keep in sync only "index" and "backups" directories on your backup client with server, you don't have to keep "bundles" everywhere.

ZFS has de-duplication features however the amount of RAM required for the lookup tables can become massive and I would generally not recommend using it for large-ish backup systems. Does this get around the massive lookup tables somehow?

Edit: The README addresses this under scalability. Apparently the overhead is much smaller than for ZFS.

For ZFS, putting in a 128GB SSD for the L2ARC solved the problem for me even though I only have 2GB of Ram for a 12TB ZFS pool.

And even if you give ZFS enough memory, its access pattern becomes so randomized that it's impossible to get decent throughput on hard drives.

Hopefully this system does not suffer from that.

> And even if you give ZFS enough memory, its access pattern becomes so randomized that it's impossible to get decent throughput on hard drives.

Any idea if the slowdown also happens for SSDs (with enough memory)?

I've never really tried it on SSD, so I can't say. I just know that I couldn't even get small data sets (sub 100 GB) to perform reasonably with dedup.

I do know that L2ARC on SSD didn't help, and ZIL never got used.

One small caveat: I don't think that "production ready", for a backup software, can be interpreted as "i'm not touching anything existing, so no harm done by using me". If it's a backup system, people will rely on it. So it's either production ready or it's not.

I went googling for a survey of backup tools and ended up at.. Hacker News: https://news.ycombinator.com/item?id=8621372

Hmm, https://www.tarsnap.com/ does this...

MMh. Test suite seems to have a lot of gaps.

-3 minus votes? wow. Never expected that.

...a backup program without tests is asking to loose your data.

Reply on news.ycombinator.com