Counting the cost
Counting the cost
While advances in storage technology have the potential to re-write the storage cost equation, nothing it seems will change the appropriateness of the saying – “it’s horses for courses” - when it comes to data storage solutions.
Common sense would say that you don’t tether a race-horse to a cart which is sitting at the back of the barn. The same rule has always applied to storage technologies. Why pay the premium of using fast, hard disk-based RAID systems for data which has not been accessed for months, or even years?
In that instance, the only answer would be “because we want to spend too much money”.
However, the announcement late last year by EMC that its new line of Centera hard disk-based solutions, called “content addressed storage”, would re-write the way organisations evaluated their storage needs has created significant debate among vendors. EMC claimed the Centera would offer a “cents-per-megabyte” solution that would become cheaper than tape.
The accepted positioning of storage solutions has been based on access times, media cost and capacity. At the top of the scale are online RAID solutions offering the fastest access times, but they were the most expensive (in terms of cents-per-megabyte). The mid-range “nearline” solutions are provided by magneto-optical libraries, which offer access times slightly slower than hard disk but cheaper storage, while at the other end tape was always the slowest, but offered the cheapest form of data storage.
When EMC stated that Centera would offer a lower cents-per-megabyte solution than tape, it was essentially re-ordering this long-standing hierarchy. But could EMCÕs announcement be, as one integrator/distributor said, a “good grab for the headlines”, or indicative of a sea-change in technology? As with most debates, the answer is somewhere in the middle.
One thing, however, is certain. More than ever, organisations need to attend to the management of their data, rather than just throwing more disks or tapes at the growing problem.
Storage vendors on either sides of the fence will offer the following example of how storage requirements are growing out of hand, largely driven by the explosion in email. For example: six recipients in an enterprise receive the same email with a 2MB attachment. This will then require 12MB of storage space, when better management could reduce that to 2MB if the attachment remained unchanged and all recipients could access the same attachment. Sounds simple in theory, but in practice it reveals many data management issues that are still being addressed.
Nonetheless, email is driving storage requirements, as is the exponential growth in electronic documents and image files, as well as e-commerce transactions.
As always, the active, structured transactional databases are stored on RAID arrays. In those critical applications, the access time to the data is paramount. Traditionally, these RAID arrays were “backed up” to tape libraries on a frequent basis.
With advances in hard disk technologies, the RAID storage vendors are proposing that the unstructured data - such as Word documents, spreadsheets, image files, etc - which may not be accessed for days, weeks or months, can now also be stored on hard disk as a permanent solution. Previously, this data would always be kept for a limited amount of time on disk or MO before being archived on tape.
”With tape, to buy a library and populate it, it costs under a cent per megabyte,” said Clive Gold, marketing director at EMC. “If you’re adding the cost of maintaining it over three years, it would work out to 2.5 cents per megabyte for popular tape drives. Although Centera today is at three, in six months time that is going to halve.”
”If you look at one person looking after a tape silo, that’s about half a cent per megabyte per year,” said Mr Gold. “There might be new generations where they lengthen the lifespan of tapes - and yes, you can keep the robotics and simply change drives - but in two or three generations, the answer is no.”
Harry Christian, marketing manager at Network Appliance, said that even if disk systems were able to compete with tape systems at the same basic cost, the entirety of an organisation’s expenditure had to be taken into account.
”Our customers are more looking at a solution over a longer period of time - not just two cents per megabyte, but over four or five years, how much would it cost to maintain the software, and how would it expand as the business grows? It’s not straight out, like fuel costing 82 cents at a service station. You can have a car that uses less fuel but breaks down every five minutes. It’s all about what is the three or four year plan,” he said.
Mr Christian said that while tape based vendors have been able to claim better upgradeability in the past, the disk-based side of the industry would now also be able to contend on this important factor of ROI.
”With traditional disk-based vendors, when an upgrade takes place, it has been what you’d call a forklift upgrade. Tape vendors are well within their rights to say what they say. In the case of Network Appliance, with people who are using our first-generation filers, they can upgrade the operating system on those older filers with new disks attached to the system. That’s one of the beauties of Network Appliance, the operating system is interchangeable,” he said.
Mr Gold said that in early 2003, EMC’s Centera would introduce backwards-compatible disk nodes with 2.5 times the capacity of existing hardware, which could be installed by simply plugging it in. Maintenance on the Centera would only require a visit by an engineer once per year, he claimed, making the maintenance cost on the device “20 to 25 per cent of the list price” for comparable tape-based systems.
”The song ‘tape is dead’ has been sung for a long time,” said Ian Selway, the online market development manager for Hewlett Packard, which offers hard disk, MO and tape solutions.
Mr Selway provided a comparison of a tape system to the five terabyte Centera system quoted in a previous issue of Image & Data Manager that was available for “under $250K”.
Essentially, the tape system is half the price, including the backup management software. For a total of just over $103K “at list price”, he said HP could offer a 50-slot, 5TB tape library using two LTO Ultrium tape drives ($78K), 50 x 100GB cartridges ($8400, or $168/cart) and a three-year, two-hour response time 24/7 maintenance program from HP ($17K).
He admitted it would also need a backup software solution for “around $30-40K” depending on the configuration. But he also asked: “What’s the cost of the application to make a Centera operational? It needs an application to pull data from the server and put it on the Centera”.
Mr Selway said tape-based systems increased in cost if “there is a very narrow back-up window”, ie, when there is limited time to archive data. With those systems, the systems are “working in parallel. Yes, they want multiple drives and lots of slots, but we’re talking about tape libraries that scale up to sixty terabytes.
”And as you scale up up in tape libraries, you find it’s still significantly cheaper (than disk-based systems).”
”There are customers who need solutions that require very fast access to data, so disk is a better solution in those instances,” he said.
He said he could foresee applications for EMC’s new Centera CAS technology, but added it was not RAID technology that can be deployed across the enterprise. “It’s a proprietary system and the only thing you can use it for is CAS; you can’t use it for anything else.
”There is certainly a role for this technology, but if the reason for it is that itÕs supposedly cheaper than tape, then that value proposition is not demonstrable,” he said.
The middle ground - called nearline solutions - has always been occupied by magneto optical (MO) or CD/DVD jukeboxes.
In a recent comparison of storage systems carried out by Sydney-based storage distributor and integrator Lynx Technologies for a large enterprise customer, it showed that for a 2.2TB system, the hard-disk system would cost $210K, a magneto-optical system was $170K, while the tape system was $50K.
Mr Bijan Assaee from Lynx Technologies said tape was “still an order of magnitude cheaper” than disk. In the above instance, the enterprise opted for the MO system after it evaluated all the requirements, not the least being the ability to archive data off-site in a secure vault, something which is not possible with RAID systems.
It this instance, the enterprise traded off the slightly slower access times from MO systems, which is said to be eight milliseconds if the platter is in the drive, compared to 5-7msec for a RAID system. If the MO library needs to load the disc, then it’s usually another 30 seconds. Tape on the other hand, is much slower. If the correct tape is in the drive it can take up to 60 seconds to access the file, while loading the tape may require anything from two minutes to two days, especially if the data is on a tape that is stored off-site in a secure vault.
It has to be said that the economic practicality or rationale of storing data which has not been accessed for months or even years on high-speed storage systems is not taken into consideration in these comparisons.
According to Mr Selway, “I don’t think that IT vendors have made it easy (for their customers) to manage the life cycle of their data”. He recalled that in the mid-‘90s, hierarchical storage management (HSM) was “all the buzz, but it never came to fruition”.
”As a result, most organisations don’t do a good job of data grooming.” HSM refers to the end-users setting up rules to maximise their storage infrastructure. For example, an HSM rule would ensure once a file is 90 days old it is automatically migrated from a RAID system to MO or tape, thereby freeing up capacity on the fastest (and most expensive) systems for the most recent data.
Mr Selway predicts that economics will drive HSM-style upgrades to storage management software that will increase the efficiency and intelligence of storage solutions. “That’s the type of enhancements we’re going to be working on for Data Protector this year,” he said. (Data Protector is the new name for Omniback, the backup software previously offered by HP.)