The debate over whether disk or tape is the better solution for backup has been going on for some time now, and it seems the answer you get typically depends on who is responding to the question.
According to many chief financial officers (CFOs), backup and disaster recovery (DR) are just like insurance policies for the business, so the least expensive method is the one they usually select. This view of data protection flies in the face of what IT operations managers deal with on a day-to-day basis. For them, protecting the organization’s critical data assets is not just an insurance policy; it’s their job. They need to assure the smooth operation, recovery and security of the applications and data that run the business, no matter what. The problem is the CFO usually wins the argument, so it all comes down to cost.
Instead of viewing the disk-versus-tape debate as a purely technical problem, I decided to look at it from a non-partisan price and performance point of view. I performed some simple Web searches on prices and performance specs for tape and disk-based backup solutions. I then charted my findings on a simple graph based on price and performance. What I found is pretty interesting.
From a purely financial standpoint, tape tends to win for smaller IT shops where the performance requirements are lower than what can be achieved with six to 10 LTO tape drives. Disk-based solutions with dedupe are better as the performance and capacity requirements increase. The results may also change based on the vendor you choose, as some vendors provide fairly inexpensive tape or disk libraries, while others require the proverbial arm and a leg. To keep this exercise simple, I used an average for performance and price in the chart below. Notice that the single node disk based VTL can provide 1024 virtual tape devices at a significant discount over buying the real thing.
Cost Comparison of Traditional Tape Versus Single Node Disk-Based VTL Price/Performance
The argument for tape
Let’s look at a small IT shop with 20TB of data to protect using common practices. (I know; isn’t it strange that 20TB can be considered a small IT shop these days?) In order to keep it simple, I will focus only on the price and performance of the drives, so let’s eliminate the media requirements from the costs and just focus on the price/performance of the drives versus a disk-based solution.
Performance requirements: Assuming 20 terabytes of production data and eight-hour backup window:
- LTO3 = 80MB/s = 288GB/h = 6.91TB day (2.3TB per 8 hours/drive = 20/2.3)= 9 drives
- LTO4 = 120MB/s = 432GB/h = 10.3TB day (3.4TB per 8 hours/drive = 20/3.4)= 6 drives
- LTO5 = 140MB/s = 504GB/h = 12.1TB day (4.0TB per 8 hours/drive = 20/4.0)= 5 drives
- LTO6 = 160MB/s = 576GB/h =13.8TB day (4.6TB per 8 hours/drive = 20/4.6)= 5 drives
To keep this an apples-to-apples comparison, we need to consider using tape library type drives versus individual tape drives, so there is no manual requirement of swapping tapes as with a disk-based library such as a virtual tape library (VTL). I found that the average price for a lower-end LTO3/4 robotic library with four drives is about $37K. The price for a four-drive LTO5 library zooms up to more than $200K. Interestingly, once you go to a library, the actual overall sustained speed drops a bit as you add in the delay of the robotics, so performance is linear to the number of drives. Let’s skip that issue to keep the graph simple and use the native drive numbers for performance.
A sampling of recent tape drive pricing from the web:
- LTO3 = $1,200
- LTO4 = $2,500
- LTO5 = $3,800 to $15,000
- LTO6 = $4,500 to $18,000
Note: The higher prices for the larger drives are for those that fit into the enterprise-level libraries, which usually start at six drives and go up to hundreds of drives. As an example, an STK SL8500 can hold up to 640 LTO drives and backup data at around 320TB per hour, but the price is quite high at that level of performance and capacity.
Tape-based solution costs based on a 20TB example with an eight-hour backup window:
- LTO3: (1200 x 9 drives ) = $10,800
- LTO4: (2500 x 6 drives) = $15,000
- LTO5: (3800 x 5 drives ) = $19,000
- LTO6: (4500 x 5 drives) = $22,500
The disk-based dedupe side
The price for disk-based dedupe solutions that compete with tape are also all over the place. They can start out at about the same list price as the average four-drive tape library, but also go up to the multiple millions in price. Although some vendors provide virtual machine-based disk appliances that are low cost and run under VMware, I will omit those for this exercise, as they are limited in overall throughput for this comparison.
Estimated disk-based backup appliance pricing
- Entry-level single node = (1600MB/s = 5.7TB/h = 136TB day) Price = $150,000
- Enterprise HA appliance = (3200MB/s = 11.4TB/h = 273TB day) Price = $238,000
- Enterprise multi-node appliance= (22,200MB/s = 80TB/h = 1920TB day) Price = $2,400,000
Clearly, the price winner in this case is the standalone nine LTO3 drives for $10,800. I’m sure that the CFO will not care that someone needs to be onsite all weekend managing tapes. Even though the prices for tape drives start out a lot more expensive in an enterprise library, when performance requirements are below a certain threshold (say 10 LTO6 drives and below) tape is the clear winner from a pure price and performance perspective. The disk based solution also enables better cost and concurrency for backup and recovery, as a single node can provide up to 1024 virtual LTO drives at much lower cost than actually buying that many physical drives.
As the chart shows, tape wins the cost curve when performance requirements are under what can be provided by six to ten LTO6 drives. Once you start getting into higher performance requirements to meet the backup window, then disk-based dedupe solutions are the better choice.
I’m sure there are some out there who could care less about the price and just want to get away from tape. But tape really is the cheaper way to go at the low end. Archives also benefit from the high capacity capabilities and low cost of tape media for long-term retention, even for large enterprises. Where tape has an issue is in its recovery time, media management costs, backup windows and total cost of ownership at the high end.
Although I gathered prices ad-hoc from the Web and made some major assumptions, I hope the comparison provides you with a useful rule of thumb when deciding between disk and tape for backup.
The good or bad news, depending on how you look at it, is that tape is not dead; it has a useful place at the low end for backup and at the high end for archives. Disk-based solutions will usually be better for faster recovery and make backup and DR more efficient and cost-effective.