In-Depth
Optical: A Better Vision for Archiving?
As you consider your archive options, don’t count out optical.
A few weeks ago at Storage Networking World Europe in Frankfurt, Germany, I reconnected with Nigel Street, group chief executive for Plasmon PLC. I first met him a few years ago on a visit to their U.S. operations center in Minneapolis.
A lot has happened since the last time we spoke. Plasmon made some acquisitions and relocated their U.S. operations to Colorado Springs, placing themselves in the heart of storage development with a brand new technology called Ultra Density Optical (UDO). UDO should have breathed new life into optical storage—raising capacity to 30 GB per platter, with 60 and 120 GB on the horizon—but it didn’t. Some folks have never heard of UDO because of the deafening noise of vendors pushing archive to tape or disk.
Despite this fact, Nigel seemed in good spirits as we chatted over coffee. He noted that acquisitions had enabled his company to build its own drive electronics. With Konica Minolta backing UDO format, and Mitsubishi also manufacturing the drive media themselves, Plasmon found itself in the enviable position of being able to supply OEMs, such as IBM and GE Healthcare, with all of the 5.25-inch format optical technology they could handle.
He noted, however, that Plasmon also learned that “if you build it, they may not come.” When UDO adoption wasn’t high and to the right, Street’s folks went back to the drawing boards. He bought RAIDTEC to obtain the technology for front-ending UDO drives with network-attachable RAID disk. They built a soup-to-nuts optical storage appliance, based on a robust Linux kernel OS and supporting both NFS and CIFS access, for use in building easily-deployed, long term archival or reference data storage. That UDO archive appliance is what Plasmon is selling today.
Street it is serious when he says that archival data ought to be stored on an archival platform. Tape and disk simply do not deliver the durability or cost-efficiency of his UDO solutions. He has some pretty compelling TCO models and white papers to help make the case.
Using a three-year/20 TB archive, Plasmon compared costs of ownership of archival options based on Linear Tape Open (LTO) tape and DVD at the low end and EMC Centera at the high end. The comparison to Centera was, I suspect, provoked by some EMC marketing materials claiming that their wares were ten times less expensive than optical solutions for data retention.
According to their independently verified analysis, “The results clearly show that the Total Cost of Ownership of LTO, DVD and UDO archives are very similar and far lower when compared to Centera. Contemporary optical technologies such as DVD and UDO remain very competitive with tape archive and are much less expensive than the Centera disk-based archive.” Looking at the associated chart, we are looking at three-year TCOs between $110 to $190K for tape and DVD, about $200K for UDO, and nearly $700K for Centera.
From an acquisition cost perspective, UDO was among the least expensive technologies given that a much lesser drive count was needed to service a 20TB repository. To achieve the desired serviceable repository size and to meet the standard workload in terms of data reads, only 4 UDO drives were needed, versus 6 DVD drives, 8 LTO drives, or 64/96 500GB Centera drives, depending on whether you wanted a parity striped or mirrored RAID configuration.
All configurations have available write-once-read-many media options, though some present additional costs for this type of media or functionality. Optical (DVD or UDO) is usually the hands-down champ in such comparisons anyway: non-repudiability of data is a “gimme,” given the static nature of optical writes in non-rewritable media. That said, optical traditionally fell short from the perspective of data deletes, which is the twin sister of data-retention requirements.
However, on this subject, Street reminded me of his “compliant WORM technology” that admirably addressed the problem of deletion. “Using our firmware-based shredding technology, you can absolutely delete data that has exceeded its retention period. Just melt the disk sectors where the data is written so that there is no data left.” Compared to data deletion on disk and tape, which leaves tell-tale under-layers of data that could be recovered long after software shredders have done their work, deleting data on optical media is clearly the more robust solution.
UDO load times are better than DVD or tape—five seconds versus 15 seconds and 19 seconds respectively—and no comparison is offered to Centera, which should present even faster data retrieval rates. However, this advantage of magnetic disk needs to be weighed against other Centera deficits . In reality, Centera’s hashing algorithms are slow and Plasmon has front-ended the UDO technology with RAID cache and can beat the Centera ingestion rates.
According to Hal Weiss, a senior systems engineer and architect for Baptist Memorial Healthcare Corporation, and a user of Centera, data load times are irrelevant if data itself cannot be found. Header hash collisions, which occur when two or more data objects have the same hash value, occur with greater frequency as Centera file repositories grow larger. When multiple files have the same MD5 hash value, none of the files can be opened. Moreover, according to Weiss, the indexing system used by Centera causes exorbitantly long index rebuild times if a disk drive fails—calculated in days or even months. Lose two disks, he says, and you lose data forever.
Centera also loses the comparison when it comes to power costs, hardware maintenance costs, and software acquisition and maintenance costs. Contrary to Hopkinton marketecture, Centera ends up costing about three times more than tape or optical solutions to the archive problem.
Street, to his credit, did not ask me to take his side in the Centera-as-archive debate. He simply provided me with an abundance of evidence that seemed to make a strong prima facie case for his position that optical rules and everything else drools. Issues we did not discuss this time around included write speeds, data format refresh requirements (how often does the format of the data need to be altered to keep it machine readable in the future), and the media cost factor skew represented by “pock-marked disk” (e.g., disks whose sectors have been deleted through melting over time). Sifting through his documents, I do not see my lingering questions regarding these issues—which go to the efficacy of any long-term archiving scheme—adequately answered.
All in all, Plasmon’s key differentiators come down to simple things. For one, optical is tried and true. There is actually 20-year-old optical media out there that can still be read. Contrast that to disk and tape, with their respective 3.8 and 11.5 Annual Failure Rates, and optical scores very high on reliability.
Second, optical technology is truly standards-based, unlike many quasi-standards that surround current tape and disk archive approaches. This means that media written with a Plasmon UDO drive should be able to be read by any UDO drive from any platform vendor. Try reading an LTO tape written in an IBM drive in an HP drive, if you want to see the difference clearly.
Third, optical professionals are not in a contest with magnetic disk professionals (unless the latter seek to deride the optical value proposition with nonsensical marketing claims). What this means is that Plasmon couldn’t care if you originally captured data onto an EMC DMX, Clarion, IBM Shark, Network Appliance Filer, HDS TagmaStore, or any other array. They can work and play well with everything out there and can compliment infrastructure you already have. Centera, on the other hand, is a hardware lock-in strategy that is usually introduced as a drag-along behind other EMC hardware purchases and contextualized as a “one stop shop” regardless of the inferiority of the solution when compared to other solutions.
Plasmon features a robust file catalog with more metadata and file differentiation than one usually gets from a tape-based archive solution. Like content addressable storage solutions, it avails itself of search and retrieve engines. However, unlike hardware CAS, it doesn’t require you to buy only brand X hardware going forward, and that is a big plus.
What I would really like to see, now that Plasmon has created an appliance product, is use of a software-only CAS engine, such as a Caringo’s CAStor, to index the data being written to UDO platters. File catalogs are great, but they also give you something else to worry about when backing up.
That’s my two cents. As you consider your archive options, don’t count out optical. Your comments are welcome: jtoigo@toigopartners.com.