In part one of this topic, I expressed surprise at people in the industry who compare the cost of disk and tape as an “either-or” proposition. In fact, the optimal solution for most is some combination of these technologies. So what is the right combination?
First, you need to think about “backup” as three different tasks and decide how important each of them is to you. They are:
1) Day in, day out backup and recovery
2) Near-term DR or site-loss protection
3) Long-term data retention
Some organizations use the same backup system to provide all three, but IT departments that think of them as separate tasks needing different technologies often improve results and reduce costs.
Backup and Restore – This is a great place for disk, especially if deduplication appliances are involved because they let users hold enough backup sets so that most, or all, of the restores are from disk. People see faster backups and much more streamlined restores. Cost factor? At acquisition, the disk systems are almost always more than tape, and conventional disk, in the amount needed, is almost always more expensive than deduplication appliances.
Near-term DR Protection – Conventional disk usually falls way short in this area. Removable media works, but the costs can be high, especially if there’s a need for frequent off-siting. Deduplication appliances with replication provide a great alternative—backup sets get moved off site every day (while a local copy is kept for restores), only new blocks are moved so bandwidth needs are low and the process is fully automated. DXi customers using this solution report 90% or more reduction in admin and dramatically lower costs for buying, transporting and storing media. Cost factor? Replicating appliances have the lowest ongoing costs with positive return on investment commonly reached in about a year. But to see the advantage, users have to factor in real costs for ongoing operational expenses.
Long-term Retention – Holding data for months and years is not an issue that everybody worries about, but if you need to hold data for extended periods, removable media (tape) is far and away the most cost effective. Tape in a vault is inexpensive, requires little power/cooling and it can provide secure off-site protection at the same time. When replicating appliances provide the backup, restore and DR protection, many users find that monthly tape creation for deep storage is adequate so they reduce direct expenses. Cost factor? Tape wins by a mile over any disk, even when deduplication is factored in. In fact, the new disk backup archive products marketed for long-term storage start out much more expensive than tape even before users realize that they probably need a second, duplicate system to provide site-loss protection.
So the key to understanding true cost comparisons between disk and tape is thinking about the different kinds of protection you need, and looking clearly at both acquisition costs and on-going expenses.
Steve,
You say “People see faster backups and much more streamlined restores”. I would like to comment to this statement.
I know of no disk equipment commonly used today, that have higher transfer data rates than current tape equipment are spec’ed. This may be not true of the infiniband disks on Oracles Exadata systems, but I am refering to “the use in most customer sites” when comparing disk to tape data transfer rates.
With hardware data compression, it is common that tape drives today to have internal data rates of 360 megabytes per second (with a 2 or 3 to one compression ratio). 120-160 megabyte native data rates with large blocksizes.
When backups (disk to tape occur), it is NOT the tape drive that limits the backup times. It is the disk drive transfer rates that limits this time.
And with restore times, if longer from tape vs disk, then in the large part, would be more to do with the application involved than the tape drive used.
I don’t agree with your performance premise.
You raise a good, and a really important point–and it’s a position that we know very at Quantum since we provide lots of backup systems based on tape drives. Tape drives are built for streaming and they have the capability of being able to send data faster than most disk systems. While that is true, it also true that a large number of our customers report significant increases in performance when they switch to disk.
One way to resolve the difference and allow both statements to be true is if we assume that the customers who see changes were not able to stream data very well and were underutilizing their drives–so the drives spent too much time stopping, starting, and repositioning. For slower and more intermittant streams, disk can be significantly faster.
That’s the view I suspect is most likely, but I’d be interested in what others have seen.
Steve,
I agree with your view of what is most likely happening as viewed by the customers today.
It is in fact the applications use of a streaming target (tape) that causes the poor performances. IE, an application issue, per say.
These tape application issues can easily be fixed, in my humbled opinion.
I contend that most tape backup applications really don’t understand how to use the advanced features of todays streaming tape drives.
The progression of tape technologies from round reels (7/9 track days) to 18 track cartridge tape and on to very high capacity “fat” streaming tape drives of today, has had a performance cost to many customers, as you have stated.
Let me explain. The start and stop times through this development process has been becoming longer and longer, with each subsequent tape drive releases. We went from a capstan motor system to move tape on the 9 trk devices (of around 1 millisecond start/stop) to a streaming system of moving tape that use a reel servo system (to as much as and even in excess of 3 second per start/stop times).
So it is an obvious statement I make in saying, Tape applications should have been changing to reduce the real affects of these start/stop “events”.
Some older tape commands and sequences that actually causes the excessive start/stops events should not be issued or issued at the correct times within the job stream to allow these new drives to continue streaming. Their are also newer tape commands that allow for better “streaming” as well. And as you have said, the excessive start/stop events IS what is causing the poor performances seen at most customer sites.
And the applications should also take better advantage of the increased interface speeds of today’s tape technolgies, by using optimum block sizes and understanding the types of work loads that really improve the customers wall clock times to complete their tape work.
I would venture to say, many many tape applications were actually written in the days of those 9 trk tape drives, and have not been changed that much, since. IBM’s DFSHSM is a great example of just this type tape application. I am sure there are many more.
So, if you believes that the answer to the performance problems with tape is to switch to disk instead, then I contend, the correct approach would be much cheaper for our customers to work with the tape applications people and their tape vendors, as a better way to solve their performance issues. Fix the applications to help improve their problem. Don’t have them go through the extreme expence and costly migrations of switching to disk as a backup device. Let’s help these customers leverage their cost of ownership in tape.
[...] [...]