Enterprise Storage: Storage Buyers: Beware the Ides of March
We all know the story. On that fateful day in 44 B.C. when Julius Caesar set off to meet with Brutus and his other protégés in the Roman Senate, a soothsayer cautioned the dictator to "beware the Ides of March." He ignored the guidance and and met the rather gory equivalent of the modern parliamentary vote of no confidence.
A little more than 2,000 years later, historians have determined that the reason that Caesar allowed himself to become the first example of a properly executed ballot – one with no dimpled or pregnant chads, just perforations – was not arrogance in the face of destiny. To the contrary, the guy probably missed the significance of the warning altogether.
In deference to all of the would-be Caesars in corporate IT departments, this month’s column offers three storage-related advisories that may be a bit more useful than the advice that the Emperor of Rome received on his way to work two millennia ago. These caveats may be helpful in sifting through the market-speak of vendors seeking to separate you from your storage budgets.
Beware the Ides of EMC
The IPS Working Group of the Internet Engineering Task Force has been hard at work for over a year on the development of an open standard for delivering an easy-to-implement SAN using TCP/IP, nicknamed "iSCSI." Some leading proponents of the iSCSI standard are IBM and Cisco Systems, though significant contributions have been made to the effort by engineers from several other vendors, as well as from academia.
The IPS Working Group is co-chaired by a senior engineer from EMC. He has done a generally good job of keeping his corporate affiliation separate from his official duties, but was compelled in mid-December to issue a statement "with his co-chair hat off" about an existing patent held by EMC covering pretty much any use of SCSI over any processor-to-processor network. Patent #5,996,024 (the "024 patent") had an application date in 1998 and potentially gave EMC ownership of significant components of protocols, such as iSCSI and Fibre Channel over IP (another standard being developed at IETF).
Many, including myself, found the notification rather chilling. The co-chair passed along the generous offer of EMC legal counsel to the effect that EMC would allow the use of its patented technology by anyone who grants to EMC a reciprocal license for the use of their technology, which encompasses the patent.
Apparently, this offer allows the patented work to be included in an IETF Request for Comment (an RFC is the rough equivalent of an open standard in IETF parliance). Earlier procedures at IETF would have immediately squelched any development effort based on "encumbered" (e.g., patented) technology in order to save a lot of people valuable time.
In response to the EMC assertion, other patents were brought to the fore, including existing patents covering SCSI over Ethernet and SCSI over AppleTalk. These were offered by their holders not to assert ownership over the work of the IPS Working Group, but to advise EMC that their "broad patents" covered much the same turf as EMC’s patent, but had earlier patent application dates. It appeared that the intent was to diffuse EMC’s claim to ownership of any or all of the work of the IPS Working Group in iSCSI and FCIP.
Off-line "assurances" have been received from many members of the Working Group that EMC will not prevail if and when it brings a patent infringement case to court in response to any vendor’s iSCSI-based products. Commentators claim that the patent is too broadly stated to provide a basis for an award.
This soothsayer recommends getting an opinion from legal counsel before settling on a strategic IP SAN technology to ensure that your IT department is insulated from what may shape up to be a real money maker for the trial lawyers. Sometimes, more than technology must be considered when endeavoring to solve technology problems.
Beware the Ides of LTO
All things considered, I like the Linear Tape Open initiative. I like the idea of a standard tape format that is supported by multiple vendors. I like the idea of having diversity of sources of supply for a storage technology that I regard as critical and strategic for my environment. And, I like the idea of taking many of the best-of-breed characteristics of different tape standards and building one really good standard that captures them all.
What I don’t like are secrets. One of the best kept secrets about LTO’s Ultrium technology (at the time of this writing, at least) is the problem some vendors are having making the darn thing work with Microsoft server platforms.
Before anyone leaps to any conclusions, this little issue did not come to my attention courtesy of LTO arch enemy, Quantum Corporation, manufacturer of rival technology and current king of the hill in tape, Digital Linear Tape (DLT). It is also not a condemnation of LTO, Ultrium or the industry effort to build and promote the standard.
It is simply the case that LTO has a problem with Windows NT Server and Windows 2000. Those operating systems transfer blocks of data that are too short to keep an LTO drive streaming. The result is extremely poor tape speeds realized from the technology when used in connection with Windows NT. You won’t read this in the LTO marketing literature.
The official response received from the LTO Initiative when queried on this issue: "There is nothing in the format that would inhibit the streaming of data. However, it is believed that the issue is related to the drive implementation of the NT operating system. In other words, the problem is most related to the NT operating system, and it is an issue that would be resolved at the product level in terms of how the drive is implemented."
One of the smartest guys I know in the tape backup software business, BakBone Software’s VP of Engineering, Fabrice Helliker, suggests that there is nothing particularly unique about the difficulty in operating Ultrium drives efficiently with NT (or any other operating system). He observes that drivers offered by operating system vendors often fail to optimize tape drive technologies, whether due to a poor understanding of the technology involved or to lack of interest on the part of developers. Conversely, hardware vendors are often constrained by a lack of access to OS software internals to write drivers that will deliver the best capabilities of their equipment.
To obtain the best performance from a tape drive, Helliker says, often application software vendors (i.e., tape backup and restore software developers) need to write their own drivers. Plus, they need to deliver the capabilities to tune software to ensure maximum drive performance.
This goes a long way toward explaining the Ultrium/NT problem, but not the silence. I would rather see the LTO folks come clean up front, explain the problem of drive performance and possibly recommend software products that can ensure optimal drive performance, than to keep the whole matter silent.
Beware the Ides of Virtualization
While we are on the subject of storage software, consider carefully this point. Storage virtualization is a prerequisite to delivering the value proposition of SANs, which is a storage infrastructure offering dynamically scalable volumes presenting data for use in a heterogeneous server and storage device network. In other words, for SANs to deliver the goods, storage devices – physical disks and partitions from different arrays – must be grouped together and virtualized and these virtual volumes must be able to be shared, absolutely and unequivocably, between servers with different operating systems.
There’s the rub: Operating systems don’t like the idea of a dynamically scalable drive. Servers want to format their disks and to assign them a volume name and drive letter. They do not understand that virtual volumes can scale and they don’t want to know about it if they do. Format a volume to 30 GB, then add 10 GB, and the operating system will continue to see 30GB.
Veritas Software is one company at the forefront of efforts to virtualize storage via software. They tout two "layers" of virtualization solutions: The first endeavors to consolidate primary disk storage and allocate it to any server via a SAN, cluster or network. The second layer enables servers to access the consolidated storage as a set of shareable, logical volumes.
From where I am sitting, their ambitious effort is useful both as a driver of solutions for volume virtualization and also as a springboard for critique for current volume virtualization technology. My hat is off to Veritas for promoting the Vertex initiative, which is among the first multi-vendor efforts to attack the Gordian Knot of virtualized volumes, and for placing themselves in the line of fire by introducing products that, in some cases, show how far we still must go before a real SAN could possibly be realized.
In lieu of an operating system capability for dynamic volume resizing, Veritas has its own utility for resizing volumes without requiring a server disk reformat and reboot. From the trenches, the word has come that the utility is pretty good at resizing volumes in a UNIX OS, not so good in the NT environment. When it doesn’t work, it is wise to have a backup of the data stored in the volume that you are trying to resize, since it may be hopelessly lost in a resize-gone-awry.
Another thorny problem about virtualization via software is data recovery. When data must be restored to disk from tape, it must filter through Veritas’ two layer virtualization scheme, potentially incurring substantial delays in the process. This is potentially bad news for those companies who have gone to great expense to backup all those critical terabytes to tape, only to realize that restoral speed may be up to 80 percent less than backup speed.
For the IT manager, it is important to be aware of the limitations of current virtualization techniques and understand their impact on important considerations like data restoral before adopting the latest and greatest virtualization technology. The data you save may be your own.
The old adage remains true: Those who forget the past are doomed to repeat it. In the Roman calendar, every month had its ides, and so it is today. The modern IT Caesars must maintain their vigilance against "marketecture" – technology placed into the hands of creative marketing forces – in order to keep strategic plans from becoming just so many shredded pages.