Online Backup Solutions?

OmnipotentEntity asks: "I'm an IT Manager (and also a lifeguard, don't ask) for a small private club. Recently parts of our server's RAID went bad just as Hurricane Dennis hit, making life a living hell for me and everyone involved. So, I figured perhaps backing up information online would make stuff like this less incredibly painful. A quick browse of Google will show that there are a lot of businesses offering automatic, offsite, online backup solutions. It seems it's becoming a big thing. The largest problem is that they all look alike -- same implementation, similar websites, it looks like someone came through this part of the Internet with a cookie cutter, and by the information available on the website and pricing (which may or may not be available without filling out 100 forms) I can't tell a good company from bad company. I've never had any experience with any of these companies, and I wanted to know if any of you guys had, and if so what were your experiences with them? What are the things to look for? What are the things to avoid? Am I barking up the wrong tree?"

Comments

Offsite Co-op?

[kwerle]

I'm curious to know if there is any kind of off-site co-op. You know - you store my data, and I'll store someone's. Encrypted, blah blah blah.

Call me a commie - but why not?

Re:Offsite Co-op?

[dgatwood]

Agreed. Online data storage is hideously expensive. A coalition of businesses and/or individuals doing a shared backup strategy via a bittorrent-like protocol would be far more effective at preventing data loss, for far less overall cost to everyone.

Of course, that sort of mechanism doesn't help if your purpose is to use backups for historical data retention, but then again, if that's your goal, online backup doesn't make sense anyway.

What would be nice would be for this sort of mechanism to be sufficiently simple that an idiot can understand it. You specify the number of unique copies (n) of your data based on how much you care about it. In exchange, you agree to store 2n times as many gigs of information for other people on your drives. That space is reserved in advance at upload time, and freed when you tell the software that the backup of that data is no longer needed.

To prevent abuse, laptops would not be allowed to participate, as the availability of data backed up on someone's laptop is dubious at best. Machines participating must have either a static IP or dynamic DNS (or, ideally, the software could automatically register some sort of free dyndns type name for you).

During the first 72 hours prior to the backup, the machine must respond to at least 75% of hourly requests for confirmation from other machines that have copies of its data. If it does not, it will be assumed to be a laptop and the data stored will be disposed of after 72 hours as space is needed. This means that you can use it if your machine is dying as a temporary backup mechanism, since the data won't go away immediately, but at the same time, will effectively prevent abusing the system by using it to backup people's laptops.

After 72 hours, the confirmation rate will decrease to once per day. A host that has been gone for more than two weeks will be assumed to have been abandoned. However, there should be a mechanism for making one machine double as a stand-in for a dead machine for an arbitrary period of time, so long as it provides enough storage to meet the original machine's obligations.

In addition to confirmation requests from the copyholder, the machine with the original data should attempt (daily) to contact each copyholder to verify that bidirectional connections are possible, thus ensuring that if the data needs to be recovered, it can be.

Obviously, since all data would be encrypted, the encyption key would be stored in a file on system being backed up. This means that you MUST back up if you ever want to recover your data....

Re:Offsite Co-op?

[Jim McCoy]

How does this sound?

You store my data, I will store yours.

Error-corrected and replicated so that 50% of the cloud could disappear and you would still have 4 or 5 nines of reliability.

Per-file, content-dependant encryption (e.g. every file gets its own AES encryption key)

Free accounts have a 10:1 provided vs. consumed ratio (to cover replication and error-correction bloat, with the ratio expected to drop over time) and people who want to buy a better ratio or even not have to provide space can do so.

Access to data backed-up by any of your systems from any other system you have installed the software on. (No more need to fiddle with system-to-system sync to make sure you have access to all of your files.)

Sound interesting? If so, head over to Allmydata and sign up for the beta test. [Windows only at the moment, but OS X and Linux versions will be available in a couple of months...]

Re:Offsite Co-op?

[dgatwood]

Periodic queries are sufficient if you do it right. If instead of two copies, you have... say five or ten, the probability of inaccessibility drops substantially. Query at most once per hour.

The traffic should be minimized through the use of what I would call "data affinity". Data from a given source should naturally tend to congregate on the same servers as other data from a given source. This, coupled with internal data checksumming on each host, means that the number of messages should be relatively small even for large amounts of stored data.

It would amount to "verify integrity of all content for host foo.bar.org", followed by a response packet: "x bytes of data, checksum 0x483957483028abcd, no bad chunks detected". When something goes wrong with a chunk, it would have a chunk list attached, and the original server would send replacements for those chunks.

If, during the normal course of operations, a chunk of data didn't checksum correctly, the server would randomly request it from its neighbors and/or the source until it found somebody who was still out there. Each data server should be able to checksum itself fairly easily.

IMHO, the only reasons for periodic queries are A. to make sure a host hasn't gone down permanently (and haning several copies means that it should be safe for detection of this condition to occur over several hours rather than minutes) and B. to prevent laptop users from putting their data out on the network and then going away without contributing back to the community by providing shared storage for everyone else.

Now there is the problem of data integrity if a new copy of data is written out to the distributed filesystem while some copies are not online. Thus, each chunk should be versioned. If a new copy of a chunk is written while a copyserver is offline, the other copyservers should tag this fact, make a new clone of the new data, and periodically try to contact it over a period of time to inform it that the data is no longer needed. After a period of time (say, two weeks), they should give up and clone any additional data that was shared with that copyserver.

Similarly, if a copyserver is brought back online after a crash, it should try to contact the other copyservers and the masterserver and ask if any of its data is still relevant. It should do this periodically, with some eventual timeout (say, two weeks).

Use gmail.

[Garridan]

Gmail gives you 2.42 gigs of storage, and growing! Never delete anything!

Do it the old fashioned way

[rerunn]

Create backups, then take them home with you if possible. Doing online backups leaves you at the mercy of the provider.

Storage Size?

[Conception]

I think online backups won't be the future for anyone. If you have a 400GB raid, and you want to back that up, we're talking a lot of time and a lot of bandwidth to transfer that to the online storage. Tape afaik is still the best way to archive data.

Gmail

[slideroll]

Some people are using Gmail as an online backup system

Good Solution

[pastpolls]

We have offices in two cities, and on top of our tape backups, we backup to each other. City one backs up via VPN and data encryption to City two, and visa versa. we are actually two seperate companies with the same parent company, so we encrypt the data (even over encrypted VPN) just to be safe from the prying eyes of people on each end.

True story: We both run Citrix servers, and one time we had a data loss at my location. Within an hour, we restored our database and application to an extra server at the remote location and used Citrix to connect our users here to the main database. I could then work on restoring from tape, without the pressure of true downtime, just inconvenience time, which I and management can tolerate.

rsync+torrent=backup_cloud

[4of12]

I was looking for a free application like that a few weeks ago and found this guy's nice write-up of desired features.

Re:Online backup? - Capacity

[HavocBMX]

Basically, it works based on sythetic backups. You have to picture that the server machines are running something like CA's ArcServe or Veritas Backup/Replication Exec 10. Both of these products use a back up methodology which although different in implementation does perform functionally realitvely equivelent things. For example take Veritas which is basically the industry standard. What they do is essentially take a snapshot of the storage and only backup those files that have changed. From those weekly snapshots they in turn generate a synthetic full snapshot. Which is then used in place of the full backup for the next week. So the only backed up files are those that have been changed. This reduces the time of backup and allows for many gigs to be backed up in a rational time frame. This in turn allows for online backup and replication.

Re:Backups online

[Tekzel]

I use external USB hard drives with 400GB storage per drive. Swap drives at locations weekly with a spair set. Fast no reaccuring costs and does the job. Tapes are too slow, online is too slow / cost can be a lot for a lot of data.

I just have to ask. How is this post off topic? Sure, it isn't about online backups specifically, but it is a very reasonable alternative TO online backups.

Some people and their children.

md5 != bit-for-bit integrity

[bofh23]

md5 or other checksums do not guarantee bit-for-bit integrity. They are just a way to gain confidence about the integrity of files without resorting to a much slower bit-for-bit comparison.

Re:You confused backups with availability.

[avronius]

Good rebuttal arete - some very valid points. My last words on the subject: Larger corporations have an easier time justifying "warm backup" sites. Streaming data (using rsync or similar) that has changed is simple enough, and usually they can afford bigger pipes. In the smaller business market, ISP's often charge based on capacity of pipeline utilized / duration of time / or some other "Penalize the greatest users" scheme (I don't think that is 100% evil, but it's not a friendly business model). While I am a strong believer in a combination of services (I'm very paranoid about losing someone else's data), if you can only afford one service, I would still recommend Tape/Disk based archival. As for how to get the tapes / disk / pieces of ivory offsite? I would recommend using a data storage company for storing your backups. Most will come to your business, bring an older set of tapes, pick up a new set of tapes, and place your tapes in a fireproof vault off of your site. I am a firm believer in the power of data transmission, but I wouldn't bet my business on it. - Avron

312's LeanOnMe did this! Safe AND secure!

[skibrian]

This has already been built. 312, Inc. did it. tell me if anyone is interested in blowing the dust off it again. We killed it November of 2004 because a lot of people didn't want to trust random people to store their encrypted files. I'd love to get this back out the door...does anyone think there would be demand? We've spent 2 years developing. Any ideas on how my company could recoup some of that investment?? A fair pricing method? The software exists...I just want to know how to unleash it, and it's been terribly frustrating as people somehow missed the value of it last year. Looking for any feedback....send an e-mail to sales@312inc[dot]com (I don't want my personal account spammed) if you want to contact me directly with ideas on how to market our product. For now, we are charging ahead with "BitVault" which is like all the models you guys are complaining about. If we can figure out how to sell LeanOnMe at a fair price...we'd do it. Thanks! www.312inc.com (google for cached websites and documents regarding our past software R&D efforts)

Re:Backups online

[Sun]

No, it does not.

http://www.lingnu.com/backup.html

You hack one server. One copy of the data gets corrupted. Second copy, however, is on a server that can only initiate outgoing connections. You cannot hack that one from outside. By the time the data gets synced, the hash proves to be wrong, and we know we were hacked. Restore from good backup, and we're done.

Shachar