Network Attached Storage For Database Applications And The Impact Of DAIS.

Filers are becoming an increasingly popular storage solution for database applications. This trend has been driven primarily by a greatly reduced total cost of ownership (TCO) when compared to direct-attached or SAN-attached storage. The Direct Access File System (DAFS) protocol promises to further accelerate this trend by adding high performance and transport independence to the TCO advantage.

Traditionally, database servers have been large complex systems configured with large amounts of memory and disk storage.

To improve the reliability and availability of these database servers, add-ons implementing some form of RAID protection have in recent years been utilized. Some of these solutions use mirroring (RAID 1), while others use parity protection (RAID 3, 4 or 5); some use hardware support, and some use special software.

However, these solutions have some shortcomings. Mirroring requires significant additional disk space. Software RAID consumes CPU cycles (reducing the CPU cycles available for database processing). Even if hardware RAID is used, CPU cycles are still required to manage the file system on which the database datafiles are placed (datafiles being those physical disk files in which a database stores data).

Database applications themselves are a software category in their own right due to their very I/O-intensive nature. Rather than simply relying on the features and capabilities of the underlying operating system to provide performance, scalability, and management, database engines have traditionally tried to manage hardware resources directly.

For example, many database installations are configured so that the database software writes directly to a "raw" device, in order to bypass the operating system's file system buffering. In most cases this results in better performance, but it also has two significant shortcomings. First, when additional disk space is required for the database operation, the steps required to increase the size of the raw disk partition are complex and time-consuming. Second, the file management tools available with the operating system for checking consistency, performing backups, etc., are not available when raw disk I/O is utilized.

In recent years, however, operating system file management has improved to such a degree that the performance advantage provided by raw I/O often no longer outweighs the administrative disadvantages, and more and more database vendors have switched to recommending the use of a file system.

A more recent development, for example, has been the use of Network Appliance's Filers, to store database datafiles, instead of directattached or SAN-attached storage arrays (See Figure).

A Filer is a high-performance high-integrity file storage appliance, which is connected to the database server using Ethernet, and uses a standard network protocol to move database information to and from the database server. As the word "appliance" implies, a Filer is a special-purpose computer designed specifically to optimize file transfer and storage while offering plug-and-play ease of use.

Filers feature integrated RAID protection, a highly optimized file system, and extensive file management capabilities, thereby reducing the complexity of the database server, which no longer has to mirror critical data files or manage a large complex file system.

Filer Advantages For Database Applications

The use of Filers to store database datafiles offers some significant advantages over the alternate approach of directattached storage or SAN-attached storage, including excellent performance, greatly improved ease of administration, and significantly reduced total cost of ownership.

Database administrators (DBAs) today spend their time on a wide range of tasks. With talent in short supply, maximizing the use of IT staff time is an important factor in operating the database, and directly relates to ownership costs.

The use of a Filer for database storage can significantly reduce the routine administrative burden on DBAs. The task of managing disk space and reliability features is kept extremely simple. Backup activities are greatly simplified, and require less downtime. And spending time and effort to architect the correct distribution of data files across separate disks is no longer required. This frees up time for them to focus on more valueadded tasks, such as database design and performance tuning. A recent report by The Input Group, entitled "Database Storage Solutions", found that with Network Appliance Filers, DBAs spend nearly 90% of their time on value-added tasks, compared with only 60% for the leading SAN-attached storage.

Filers greatly simplify the ability to scale up disk space. Since a Filer provides one large file system, it is simply a matter of adding a new disk drive to the storage array, issuing a couple of simple commands, and waiting long enough for the filer to zero out the new disk drive. Database files can now be either manually extended or allowed to automatically extend into the new space. The Input report found that SAN-attached storage typically takes four hours to scale up by 200GB, compared with less than 10 minutes for NetApp Filers.

Filers have a unique capability known as Snapshots where the current state of the file system is preserved for a specified interval, or until deliberately terminated. Snapshots provide a read-only view that can be used as checkpoints (to recover accidentally deleted files), or for making backups. Combining Snapshots with the online backup functionality provided in most databases, Filers can maximize the availability of the database even during backup procedures. Consequently, routine backup and recovery tasks, which typically consume 35% of a DBA's time with SAN-attached storage, take only 10% for Filer environments.

Total Cost Of Ownership Advantage

The Input report on "Database Storage Solutions" concluded with a Total Cost of Ownership calculation, taking account of:

* Product cost for the underlying hardware and database tools, including maintenance costs and upgrade costs;

* Database implementation costs, including training costs;

* Operation costs (including support and helpdesk); and

* Cost of planned and unplanned downtime.

They found the annual total cost of ownership per Gigabyte for database environments is 75% lower for NetApp Filers than for competing SAN-attached storage.

DAFS And Database Applications

The Direct Access File System (DAFS) protocol promises to deliver the TCO advantages demonstrated by Network Appliance Filers to SAN and machine room environments, with significant database performance enhancements. DAFS is transport-agnostic and will run on Ethernet, Fibre Channel, or InfiniBand and fabrics.

DAFS is a file access protocol based on NFS, which is being specifically designed to take advantage of the "Direct" (RDMA) and asynchronous I/O features of standard memory-to-memory interconnects such as VI and InfiniBand. This translates to wire-speed transfers over gigabit and 10 gigabit networks, combined with a significant reduction in the client-side overhead. When the "client" is a database server, this means a significant increase in the percentage of CPU cycles available for processing database operations.

DAFS gives applications direct access to transport resources, bypassing the OS, and enabling them to optimally manage their own I/O. This model is an extremely good fit for I/O-intensive applications such as database engines. And in fact, the leading database vendors have been very active in the development of the DAFS protocol spec to ensure that it provides the features that will enable them to realize significant database performance gains.

DAFS can be implemented in a number of different ways. Kernel implementations can provide transparent access to DAFS by intercepting standard file I/O calls or device I/O calls. User-space DAFS library implementations can provide optimal performance by providing an API which enables applications to directly manage I/O. Again the leading database vendors are active within the DAFS Collaborative helping to ensure that the DAFS API Specification includes the features they need.

The DAFS Collaborative

The DAFS Collaborative was formed in June 2000 by Network Appliance, Intel, and other leading systems and storage networking vendors, with the goal of making the Direct Access File System protocol available to the industry. Today more than 70 companies are contributing to the development of the DAFS Protocol Specification and the DAFS API Specification. The group is soliciting industry review and feedback before submitting the Specifications to the appropriate standards bodies.

Additional participants are welcome and can join online. For more information, contact Werner Glinka, executive director of the DAFS Collaborative, at: [e-mail].

David Dale is an industry evangelist at Network Appliance (Waltham, MA) and a co-chair of the DAFS Collaborative.