An Abstract Art

Abstractart? Well, it’s often hard for me to understand. But the industry trend towardstorage abstraction is a different matter. This is an important development instorage and I believe vendors are starting to bring some useful products tomarket that usefully implement abstraction.

In nativeoperating systems, applications and users access storage using physical devicenames, a strategy sometimes called hard addressing. As storage systems grow insize, hard addressing makes device independence and management nearlyimpossible. For those of us familiar with drive mapping under Microsoft’sWindows, there’s the familiar C: drive. Unix users often see references todevices such as /dev/dsk/c0t1d0s6.

What weneed is a middle layer that hides the physical drive names from ourapplications. A utility should be able to access the “System Cache” or the“Public Exchange Drive” as storage without having to know the physical drivename or its properties. The benefits of this middle layer include isolating theenterprise’s applications from the underlying physical devices, improving onthe manageability and maintainability of systems, the ability to expand storageon the fly, and the potential to mix and match a variety of storage deviceswithout affecting the programs that use them.

Savvyadministrators of Windows 2000 know that a logical disk manager (LDM) was addedto the operating system to add reliability, fault tolerance, and the ability toencapsulate volume information so that disks can be easily moved. The LDM isthe core tool for volume creation, deletion, and configuration in Windows 2000.

Unixadministrators usually find a volume manager built into the operating system.Others take advantage of third party volume managers like Veritas’ Volume Manager.The volume manager tends to be a system specific utility -- that is, it workson the disk volumes for an individual server. What’s often missing is theability to span and use disks from a variety of servers.

WhenMicrosoft’s Distributed File System (DFS) first appeared, it became possible tolink storage from a variety of systems together in a single view. DFS makes itpossible for a manager to make distributed storage seem as if all the contentsof the file system reside on a single server. That’s useful, but DFS doesn’tsupport heterogeneous environments very easily. Worse yet, if you depend onlarge-scale storage area networks, DFS doesn’t adapt to the new style ofnetworked storage. Simple management tools, like disk quotas, also cannot be extendedto DFS volumes. DFS isn’t the only distributed file system, but the challengesMicrosoft faces with the technology point toward the solutions storagearchitects need.

What’semerging for system architects is a solution that shares different file systemsand volumes across a range of operating systems. These devices, called SANappliances or virtual storage servers, use a specialized server with separatestorage management software to provide abstraction. Storage appliances canprovide volume management tools, RAID, mirroring, and other managementfunctions on top of the ability to provide shared storage access.

Byabstracting the storage this way, we can allow existing storage to be migratedinto a SAN or RAID environment by taking it out of server-specific systems andplacing it behind a storage server attached to applications servers using FibreChannel or SCSI interfaces. This makes it possible to do the abstraction andfile system sharing in cases where volume managers don’t exist, and it alsoavoids the problem of being dependent on a specific vendor.

A recentsurvey shows that more than 80 percent of Fortune 1,000 companies have investedin virtual storage devices. That’s an intriguing figure, but it probably refersto network attached storage and storage area networks that provideserver-centric storage for a particular operating system. That’s fine, but whatwe really need is a solution where distributed volume managers and distributedfile systems can work with intelligent storage devices to provide true,abstracted storage.

That’s whatI expect to see happen in the near future. We will have the ability to combinethe performance and flexibility of Fibre Channel storage area networks with volumemanagers on local servers. That will mean we can share data and storage accessacross multiple systems over a variety of connection fabrics -- including FibreChannel and other SAN technologies.

There’sstill work to be done to reach this goal, but every successive announcementfrom key vendors such as Compaq, EMC, Veritas, and HP give me hope that the dayof truly abstracted storage is not far off. In addition, the Storage NetworkingIndustry Association continues to work on defining vendor neutral storagemanagement interfaces. When they’re finished we’ll be closer to understandingabstracted storage than understanding abstract art. --Mark McFadden is a consultant and is communications director for theCommercial Internet eXchange (Washington). Contact him at