Recently I was introduced to a backup application called Arq, by Haystack Software. This is a $29 app that gives you a clean interface to Amazon’s S3 or Glacier servers, allowing you to use Amazon as a backup service. Since Glacier is only $.01 per GB per month (about $10 per TB), it’s a pretty good deal. There are initial upload and then retrieval charges to consider as well, but the peace of mind of online/cloud backup is hard to put a price on.

Granted, you can buy a 3TB USB 3 hard drive today for just $130 [Amazon.com link] and ship that to a friend on the other side of the country for safe keeping, but by now I think we all know the advantages of automated, offsite backup.

Backup Bouncer Test

Anyway, this article isn’t about Arq or Glacier. It’s about a disturbing statement I read on the Haystack website, which I immediately challenged Backblaze on. As you know I’m a huge supporter of Backblaze (having written a very popular post on the topic “Cloud Backup; Backblaze in the Real World” last year), so seeing anything negative about a service I rely on is sure to get my hackles up!

The statement in question is under the header “Accuracy”, around the middle of the Arq info page. It states that Backblaze failed 19 out of 20 tests using a test suite called “Backup Bouncer”. In fact, the list goes on to show that Carbonite failed 20 out of 20, Dropbox failed 19 of 20, and so-on. Disturbing numbers, to say the least!

In complete fairness to Haystack Software, the folks who wrote this article, they did state “What do these results mean? For most scenarios, probably nothing. Any of those backup apps can restore your file contents — photos, Office docs, music files. You’ll still be able to view your restored photos, edit your restored Office docs, play your restored music. But the dates on the files might not be correct, for instance.” But regardless; 19 out of 20 failures does not instill confidence.

Backblaze’s response

So naturally, I asked Backblaze, and am quoting their response below. The short version is… chill out. It’s all good.

From Backblaze spport:

“Backup Bouncer, for the most part, is a “file system cloner bouncer” and is not a great way of testing a backup service. Backblaze is not a drive cloner, and we should not be used as such as you cannot boot from a Backblaze backup. This is by design. Backblaze explicitly is for data and specifically refuses to backup system files and the more esoteric metadata, such as Finder comments, bootability flags, ACLs, etc. Backblaze is to keep real-world data files like pictures and movies and music and Quicken files safe—standard stuff that standard users deal with. Backblaze also leans towards cross platform: you can backup a Mac and restore to a PC and vice versa.

“To be clear, Backblaze specifically backs up the Macintosh resource forks, the Finder info of “type” and “creator”, and the created and modified dates of a file. Those are the most commonly used and viewed types of metadata, and what we focus on.”

What they are saying is that Backblaze was never designed to make a mirror or clone of your system, which is what Backup Bouncer looks for. So using a clone-checker to check a non-clone backup isn’t an accurate test. The results are misleading, and frankly, dangerous. It’s not fair to Backblaze or to their users to panic people into thinking that their backups may not be safe.

Does this mean Backblaze is infallible? No, of course not. Nothing is perfect, and this is why we have multiple backups. Remember, “if it doesn’t exist in three places, it doesn’t exist”. Backblaze will continue to be my online backup service of choice.

My own test

I decided to run a small real-world test, and restore roughly 35 GB of backed up photos (about 1.5% of my 2.3 TB photo archive on Backblaze). The process was simple; I logged onto Backblaze, selected a couple of whole-month folders from my Aperture Masters collection, and clicked restore. A short time later I received an email that my zip was ready to download, and when I realized I had 35 GB to pull, I downloaded the “Backblaze Downloader” app, which allows you to pause your backup mid-go. That evening I kicked it off for an overnight download.

Results? Perfectly fine. No I didn’t open every single file (there were lots of them), but I opened several selected randomly, of different formats (JPG, TIF, ORF) and all opened fine. For the record, while I’ve never had the need to restore a massive Aperture library, I have created smaller test libraries, backed them up and restored them with no problem.

Cost of Arq and Glacier

Just to do the math; if I were to backup just my Aperture originals (masters), that’s currently 2.32 TB, or 2,376 GB. At 1¢ per GB, that’s $23.76 per month. There is also a 5¢ per 1,000 requests “request” fee to Glacier, which the Arq website describes as “approximately $.05/GB in ‘per-request’ fees”, so I think means that it’d cost $118.80 for the initial backup, although that’s not really clear to me. If someone else understands these request fees more clearly (of the 1,000 requests, is it 1 request per MB, or per file, or what?), please feel free to clarify in the comments below.

Glacier is considerably more expensive than Backblaze at only $3.96 per month for unlimited data, but if you need a perfect clone, then it’s a great solution. And let’s be honest… having yet another online backup to your precious photos can only be a good thing.