21st Century Storage.

SURVIVING THE LEADING EDGE

If you have been living in a cave in the most remote reaches of the earth, then you may be unaware of the "E"-volution. As demonstrated by the on-line retailers who could not place orders this past holiday season because they lacked the infrastructure, storage availability is central to a company's competitive position. This data intensive business model has driven double and triple digit growth rates in storage capacity whether the electronic commerce is business-to-business or business-to-consumer.

But Storage Doesn't Impact Me. Well, Maybe It Does

Of course, if you are a distributed systems manager or administrator, you probably do not have anyone in your organization with the title of storage administrator. That is one of those distasteful mainframe names that you heard existed in the dark ages. You are known by the operating system, application, or database that you manage. Storage is simply a hard drive that can be upgraded or added at will. The most pressing problems you face have been elsewhere. Distributed systems storage management only means backing up a hard drive or disk array to another disk or to tape. Plus, your company probably has a policy mandating users to do this for their own workstations.

Given that storage expenditures have been the largest single budget item for IT organizations even before all those "e- "words made application availability a make or break proposition, recognition of the importance of storage is only logical. The fundamental fact that system performance, as measured by end-user response time, is largely a function of an I/O to a disk is not news. While it is true that you old time storage administrators already know a lot about networks that attach to storage, Storage Area Networks (SANs) are going to add some new challenges.

Total Cost of Ownership

Without exception, industry analysts concur that management is the largest contributor to the total cost of ownership of even the most expensive mass storage devices. With the decline in fully configured prices per MB, it stands to reason that non-declining costs would assume a bigger piece of the cost pie, but there is much more to it than that. Management costs are climbing along with the capacity. A simplified example makes this clear: your PC or laptop.

You can (not that most of us really do) back everything up en masse and then do incremental back-ups on a regular basis. You may eventually delete old files, if you get the "Insufficient Disk Space" message when you try to save something, or you could use your personal folder on the network drive. You know what each file is because you named them, right? You can search through the hard drive(s), your network folders, and archives, if you do forget what you named it or where you put it, and can find it eventually, unless you deleted it by accident. Try adding up how much time you really do spend doing this type of thing just for a week. The results may startle you. Now, take an enterprise with hundreds of servers, thousands of workstations, a dozen or more critical applications, and 10TB of data spread across these platforms and at least three flavors of mass storage devices. Forget for a moment that some of these devices are host-attached and that the shared ones will have additional issues and consider the f ollowing questions:

* Do you know where your storage is?

* How much is being used over time?

* Can you do it without deploying the "sneaker net" to obtain the data, writing your own customs scripts, or licensing expensive and complex statistical packages that require dedicated, trained specialists?

* Do you know when you are running out of space and need new capacity?

* Do you know exactly where you have empty or wasted space?

* Ever tried to redeploy existing capacity? Do you always just buy more when your performance-modeling tool recommends adding storage to improve response time?

Answering each one of these questions can take considerable time and costs without an enterprise-wide management strategy for storage. So, most companies just throw more hardware at the problem, which simply supplies additional fuel to the fire, as they say.

Returning to the backup example, how do you get all that data protected? You can do traditional backup processing or you can duplicate the disk space so that you can have instant copies. The former can take more time than you have and the latter means multiplying hardware again. With the latter you can reduce disruptions for minor failures because you basically have "hot spares." Throw Disaster Recovery (DR) into the picture and, with either solution, you have something akin to looking for a toothpick in a pile of moving boxes all labeled "household items." Yes, the data and the applications are "out there," but how many hours or days can the average company wait to get even the most critical applications back in useable form? You actually do not have a viable DR strategy if you are focusing on file level restores rather than application recovery, regardless of whether or not it's a hardware or software implementation.

Ensuring Availability

With mass storage devices and even SANs, there has been an assumption that just throwing more and smarter hardware at a problem was the most efficient solution for business availability. Countless times, this has proven not to be the case, yet the myth is immune. Take the various large capacity RAID boxes that can be divided up between different operating systems. The catch here is that the user cannot change the amount of disk given to each of the operating systems easily. You will have a great deal of unused space in the part assigned to a group of servers that aren't growing much, while another group with an expanding application is running out of space. That happens when you have no way of monitoring and forecasting storage resource usage.

Another classic case is having two heavily accessed files on the same physical device. It does not matter how fast or big that device is; neither file will perform well, if competing for the same connections.

Take the case of the smart device that can phone home like E.T. when the box detects a problem. It's great that the problem can be handled remotely, but what if it impacted data integrity, compromising both the original and the "hot spare?" Even if this "bad data" is used for only a few minutes before the administrator is told that the problem had occurred and was fixed; the damage to the data has been done. Classic local event management is still needed to truly manage the solution.

The bottom line is that availability is not dependent upon the total size of the storage device or how. "smart" it may be, but on how well you manage the device: space, performance, and events from the various logical and physical perspectives. In other words, you need enterprise storage management software too.

Storage Area Networks (SANs)

Now, we have the 21st century solution to storage for the enterprise, the SAN. This Local Area Network dedicated to storage is supposed to solve all problems. All you have to do is configure the fabric, interconnects, and zones and everything will be available all the time to every user. Sounds good, but you still need tools to determine the initial configuration, as well as monitor and manage all those different elements and their relationships over time.

It is true that a SAN will have performance advantages for the enterprise as a whole because it will isolate one of the critical but resource intensive storage management processes, namely back-up. Also, it provides users with multiple paths to the data of interest if the network is configured properly and can transfer large blocks of data at high speed to wherever it is needed.

Fibre Channel is supposed to eliminate all performance problems that result from the fact that network bandwidth is of finite capacity. From a storage perspective, the rule of thumb has always been that space usage will increase to fill whatever capacity is made available to it. So far, the same has been proven true for network capacity, as well.

Even with warp drives, basic storage resource management is essential. Otherwise, you will only crash harder and burn brighter. Where key files are located, who is using which devices, what are the possible access patterns, where is the real workload, and where should new files be allocated are still important questions. A SAN doesn't eliminate any of these, but adds new ones, e.g. zoning.

Explosive Data And SAN Benefits

One thing that a SAN will do is expand the ability of users to create more data. With more people having more access to more data along with the ability to do more with it, storage resource availability requirements will change. As a result, there will be an increasing need to monitor and deliver service from application-related perspectives. Business-relevant groupings will become increasingly important. Effective management of storage costs is but one reason. Ensuring availability is the critical requirement in determining the real costs and benefits associated with a SAN. In an environment that fosters data growth, recovery will become even more difficult. The point is that the mass backup of files is still just that. Granted, with enough bandwidth or enough mirrored capacity, you can get it all backed up, but remember that recovery is another story.

So, in addition to continuing to manage storage resources, SANs are going to drive the demand for intelligent backup. This means that the backup process includes only necessary data and the restoration is guided and synchronized to account for the distributed nature of data and applications. It's also highly automated and self-monitoring. In other words, once the automated restoration is complete, even manual clean up that is still required is guided for administrators. This ensures that applications can be restored to service efficiently.

Reality Check

Of course, the ultimate promise of SANs involves true file sharing across platforms and real any-to-any connectivity, i.e. hooking any vendor's hardware together. All the hardware vendors speak as if this will be available tomorrow and that they are leading the charge for the development of open systems standards, but no one has nailed the month, day, and year. In the meantime, there will be a great deal of activity and competition among manufacturers, as well as among software vendors to deliver first. In addition, the promise of openness just may force all the players to become as open as their marketing literature claims. Given the sheer number of different technologies and vendors required to field a SAN, the establishment of standards might not be the result of one competitor dispatching all the rest. That alone would be a good thing for the user community.

In the meantime, here is some advice to help you sort through all of the propaganda:

* If a press release is not announcing the delivery of a product or function, ignore it.

* Architecture over Marketecture. Translation: white papers are nice, but technical specifications are a lot better.

* Selecting hardware is not like getting married. You are not looking for monogamy.

* If you buy something based on promises of "futures," get them in writing or don't get it.

* Invest in professional services to help you define your needs and make them the same people who implement, train your people, and come back in a year for a review.

* Everything has to be managed or it will cost you.

* Automate everything that you can or you will never have time to do anything else.

* Hardware (microcode) is only "so smart." Even hardware vendors are adding software for dynamic, flexible management, as well as to replace revenue lost from falling hardware prices.

* First isn't always the best. While being on the leading edge has its rewards, remember you are only a nanosecond away from the "bleeding edge."

Beverly Casstevens is the senior product marketing manager of storage solutions at BMC Software, Inc. (Houston, TX).