Database and storage management: new storage-management products reduce administration and accelerate transactions - Storage Networking

Computer Technology Review, Jan, 2003 by Calvin Hsu

The successful operation of an enterprise database depends on more than just the robustness of the software and the intuition and skill of database administrators (DBAs). Table structures, software design, processor speed and reliability play a large role in delivering continuous, high-speed database behavior, but the underlying storage performance, storage bandwidth and storage management practices have equal importance. Functional managers depend on DBAs to keep their systems running, but often underestimate the amount of time spent on storage-related activities to support the database.

Many DBAs spend much of their day configuring, allocating, requesting, monitoring, tuning, moving and updating storage to support their base disk requirements. It's a tedious collaborative effort involving several members of the IT staff. Given the choice, most line-of-business managers would rather have their DBAs concentrate on data quality, analysis and other service-level considerations instead of around-the-clock disk management.

Physical disk limitations and operating system shortcomings are to blame. Whether it's a simple JBOD directly attached to a server, or a storage area network using the latest intelligent storage subsystems, balancing 110 bandwidth, reliability and response times requires undue attention. The wrong decisions could not only jeopardize the performance and productivity of the databases, but also the availability of the data. To date, hardware constraints have put the DBA in a precarious position, with few tools to succeed.

Fortunately, a recent emergence of management products with automated features, like DataCore's SANsymphony, are helping to alleviate the storage burden, enabling application owners and DBAs to concentrate on the "business end" of their information. The most comprehensive of the new storage management offerings come as independent network nodes (not embedded in storage hardware or loaded on application servers), with a range of consolidated features that span multiple servers and storage devices.

There are five significant attributes of storage from the database perspective; the latest software addresses all of these needs:

* Performance

* Availability

* Scalability

* Disaster Recovery

* Upgrades

Boosting Database Performance

Productivity and profitability depend on performance, so the pressure is on to deliver the best at the lowest cost. Achieving high transaction rates is made difficult by the current disparity between server and storage performance. Server processors, memory and bandwidth are getting faster and cheaper. Disks are getting bigger and cheaper--but as a result of the capacity crammed on each spindle, they are effectively getting slower. The number of I/Os per gigabytes of disk space has steadily dropped over the past six years.

So what happens? One of the workarounds commonly practiced (and advocated by database vendors) is to distribute data tables and reference files across many smaller, dedicated disks. Such "striping" increases 110 parallelism and shortens access times. With it comes drastically underutilized capacity; spindles often going more than 70% unoccupied. The other problem is that no matter whether the storage is direct-attached or networked, the administrators must manage direct relationships between each table and a specific physical disk, greatly increasing the management complexity.

One successful approach to improving performance is the use of storage network management devices that add a layer of cache in front of storage. One major storage and systems supplier tested methods of improving performance in various database environments. They used DataCore's storage network management software that provided performance-boosting cache as well as storage virtualization and automation. In their preliminary tests, they achieved 85% improvement in data access performance from just a single instance of the software. Even database updates benefited by 15%.

Community Health Network of Indiana is demonstrating similar database performance gains in a production environment, with similar hardware and the same storage management software. In its case, with multiple nodes providing additional cache, it is actually experiencing a 300% improvement in database operations. Rick Copple, CTO at Community, said, "The improvement in performance is so dramatic that our doctors noticed it immediately."

Keep Databases Churning

The value of non-stop access to data cannot be underestimated. Here are some common industry statistics that reinforce how costly downtime can be:

DBAs and storage administrators work hard to minimize the impact of unexpected failures, using clustered servers, redundant paths through redundant switches to RAID arrays on the backend. However, some sources of downtime are overlooked, such as planned downtime for storage maintenance and capacity re-allocation. In a study of ecommerce websites, Aberdeen Group estimated that about 30% of all user-perceived Web server outages are caused by planned storage reconfiguration.