Tape has been the chosen form of backup media for decades, but with newer technologies, this is changing. Now other contenders are creeping up on this venerable format. What is the current sweet spot in backup media types for 2009? Is it:
- Tape
- DVD
- Blu-Ray
- Hard Drives
or something else? The basis of this recommendation is done with the following criteria:
- The media should ideally be able to hold a single full backup of the data you're working on, although...
- You can have up to 2 media for each backup cycle.
- The current working size of the data set is less than 1 terabyte. Total on-line storage capacity is less than 3 terabytes.
- Data compression is enabled for the backup process.
- The backup media will be part of a rotating set, which means the media will see re-use at some point in the year.
- The rotation cycle is fairly short, less than 30 days.
- Media is taken daily off-site. Yearly backups are taken off-site permanently.
- The backup process has approx. a 4-6 hour window to complete, although it can run over if needed.
For the record, my work currently uses tray-loaded hot-swap eSATA enclosures for 500Gb hard drives, although I see that people are still using tape, which is what lead to this question.
This is a follow-up to Evan's questions:
- What's the offsite rotation schedule (weekly, daily, multiple times / day, etc)?
As above the prior night's backup is taken off-site, and returned the next morning. At the end of each fiscal calendar, a complete backup is taken and shipped to off-site storage.
- Do you have any idea of the size of a hypothetical daily incremental backup (ie. size of the daily working set - you say "working size", but I believe you're referring to the entire backup corpus)?
Correct, the corpus is less than 1 Tb. Daily changes constitute about 200Gb.
- You say "You can have up to 2-3 media for each backup cycle"-- I'm taking that to mean that a "full backup" should use 2-3 media, not that you want to have 2-3 sets for each "generation" (grandfather, father, son, etc).
Also correct, the target is 1 "media unit" per backup cycle, although in some cases it can be extended to two.
Like anything in business, this comes down to requirements and cost-effectiveness. It depends(tm).
Here we go again! >smile< You'll end up with religious arguments in this post, if it goes the way that most of the posts about backup have on Server Fault.
You'll have the curmudgenly old guys like me who still generally recommend tape versus the trendy young guys who want to use disks like they were tape cartridges. Someone will bring up long-term retention and the longevity of tape, and someone else will chime in about how they have some IDE hard drives from 1992 that still work great.
After that, someone will mention the cost of tape media being less, per GB, than hard disk drives. Someone else will point to a weekly NewEgg special on 1TB hard disk drives and say that tape is more expensive. Someone else will factor in the cost of the tape drive and calculate the "break even" point for tape.
(No one usually argues for optical media, but I suppose there's a chance someone might.)
Personally, I wouldn't trust disks for long term archiving. You could use disks like tape cartridges (that Dell RD1000 that Russ Warren mentons is just 2.5" SATA drives inside a plastic enclosure that makes them seem "tape like" and, no doubt, is built to withstand some abuse), but you should think about the cost per media and the conditions in storage and transport.
Edit:
I've done a little spreadsheet (available at http://mx02.wellbury.com/misc/20090713-Server_Fault_Backup_Roundup.xls) that compares the following (with their calculated 1st year cost including drives):
I assumed a 5 day / week, 5 week rotation (35 days until a tape comes back around), running "full" backups with compression every day. I included the 500GB eSATA and RD1000 drives even though it was unclear if they'd actually hold the backup corpus or not.
I didn't factor in any kind of eSATA enclosures into my pricing. Realistically, there would need to be something surrounding the disks, but that's so subjective that I decided not to even bother. Handling those disks "bare" is asking for static electricity-induced damage to the circuit boards.
It's unclear what to say for a media replacement strategy. The SATA drives are warranted for 3 years (Hitachi), but I don't how they'd hold up to this kind of use. The LTO-4 tapes are lifetime warranted and typically good for 200 - 250 full passes (which would be over 19 years of use in this scenario). I have no idea what to say about media replacement on the RD1000's.
Those little 500GB 2.5" SATA drives in plastic boxes (aka RD1000 cartridges) at $599.00 ea. from Dell are a bit pricey, especially comapred to $50.00 500GB SATA drives or $41.00 LTO-4 tapes!
We still use tape for offsite. This is for backup sets up to about 10TB in size. The backup sets do span tapes, but with decent software (we have no issues with either Netbackup or NetVault, depending on the customer) this isn't an issue.
Our current media of choice is LTO-4 for new installations, although depending on the age of the installation we still have some LTO-2/3 and AIT-2/3 units (obviously the AIT units are much smaller data sets).
For your case, I'd consider LTO-4 drives, or a small robot like a HP-SuperLoader with a LTO-4 drive in it. Tapes shouldn't run you more than $100 a cartridge.
It applies to all backup media, but if you are using dual media for every backup, then buy your stuff from 2 different manufacturers and never use media from the same batch for a job.
Hard to do with some tape drives, but as HD's are more susceptible to knocks, i like to use 1TB samsung spinpoints coupled with 1TB seagates. Means that if i do get stuffed with a bad batch and yes it can happen after the backup's been verified then at least i have a fighting chance.
M.
Look at something like the Dell RD1000. You can get a rack-mountable drive system and multiple removable hard drives to take offsite and complete your rotating set. I think they go up to 500gb & 1tb compressed per drive.
As for optical media:
I've not used it, but I know that the girls & guys from DAX Archiving test a lot of optical media before using it in, for example, their Blu-ray archiving robots. I don't think the primary use for those machines is backup (though they obviously could be installed off-site if the network connection allows for that), but: not all optical media are equal, and brand and type does matter.
(They provide some whitepapers, which I have not read either. Like one written by Verbatim, claiming More than 100 years projected lifetime for DVD-R General.)
Adrian Wong of Tech ARP by accident found out that "CD-Rs that were just 7-9 years old were failing at a significant rate". Apparently in September a start-up named Millenniata "will release a new archive disk technology to preserve data at room temperature for 1,000 years", called the Millennial Disc.
Nothing beats the performance/cost/reliability of a plain old disk. :)
To guard against bitrot - use a filesystem with data checksum.
Tape has write protect tabs. Useful for protecting against a certain class of errors.
Really, it's just a matter of taking the strengths and weaknesses of each media (cost, reliability, throughput, re-usability, etc.), and choosing the one that best fits your backup requirements.