Hacker News Reader

Comment by jgrahamc | Hacker News Reader

This is why you use tarsnap. It's efficient and any trouble you can talk to Colin.

I'm a happy tarsnap user, but there are real reasons not to use it: the bus factor is 1, restores of large amounts of data are slowish, and data is stored at S3-ish prices - which is to say, not cheap if you're storing terabytes of data.

But tarsnap is indeed excellent.

While we use tarsnap for our server backups, it's not something I'd use for frequent desktop backups.

The main reason is that it doesn't have a daemon that watches file changes to keep a dirty list so it'd basically DDoS your desktop every time the cronjob ran to run the backup. Also for a desktop it's nice to have an interface where you can see the differences between backups.

The I/O thing is actually a problem on servers too. See if you can spot the point in this graph where cron.daily runs:

https://skitch.com/scotchi/gjbyc/generate-graph

Have you tried using the --disk-pause option? That should allow you to slow Tarsnap down so that it doesn't eat all your I/Os. (For desktop systems with limited internet bandwidth, the --maxbw-rate option is probably worth investigating too, unless you want your web browsing to slow down significantly.)

Excuse my presumption if you already know about this, but there are ways to make the I/O less aggressive for processes that hit the disk hard.

Since your graph header says 'linode', I'm assuming your servers all run Linux, and you should take a look at ionice[1]. It adjusts the IO priority of a process like 'nice' adjusts CPU priority. For example, prefixing your call like this: "ionice -c 3 tarsnap...." will give tarsnap the lowest priority access to the drive.

[1] http://linux.die.net/man/1/ionice

Plus, you use tarsnap in the same way as 'tar' which you're already familiar with. Backups are securely encrypted, and encryption happens on the fly. And all that goodness on the commandline!

The reasons I don't use tarsnap:

1. Not a big enough company. Colin might get hit by a bus.

2. Rackspace has no incoming bandwidth charges on their cloud.

Reply on news.ycombinator.com