Home >> White Papers >> FlashDisk Configuration Tips

FlashDisk Configuration Tips

Introduction
Winchester Systems FlashDisk RAID arrays offer a tantalizing promise of reliability and trouble free operation. It is important to understand how these benefits are derived and, more importantly, what level of protection that each mode provides.

Data Integrity
FlashDisk RAID arrays offer protection from single drive failures by recording parity information used to reconstruct lost data. RAID 3 and RAID 5 each use one extra disk drive per array to record parity information that can be used to reconstruct data from the remaining disk drives. RAID 3 stores the parity all on one disk whereas RAID 5 stripes the parity across all the member drives of the logical drive. If a four drive RAID array contains data A, B & C, then the parity information D = A + B + C is also recorded. If drive C fails, then C can be reconstructed easily since C = D - A - B. This reconstruction occurs block by block until drive C is completely rebuilt. In uncached RAID arrays, parity calculations create significant overhead. With FlashDisk, the cache eliminates this overhead and the array runs as much as 50 times faster.

RAID 1+0 arrays keep a complete duplicate striped set of your data. If one drive fails, the other "mirror" drive continues without interruption. The failed drive is replaced and completely rebuilt through a copy of the data from the corresponding "mirror" drive. This is actually one of the best RAID solutions for high availability, but requires 100% higher cost of disk drives making it appropriate only for the most critical applications.

RAID 3 and RAID 5 systems require only one parity drive for the entire array of typically 8 to 12 disks. Storage overhead is typically reduced to 8 - 12% rather than 100% making RAID economically feasible in more environments. Data integrity is preserved in any RAID 1+0, 3, 5 system by avoiding data loss due to a single drive failure.

High Availability
Drives are not the only components that fail. Controllers, buses, servers, fans, power supplies all fail. The higher your cost of downtime, the more precautions you must take. You need to look at your server environment and take the necessary steps to minimize the chances for downtime. Availability is increased through redundancy -- and redundancy costs money! So, here are the two key questions: How much does downtime cost you per hour? And, how much are you willing to spend now to reduce the chances of downtime later? Once you answer these questions, there are many high availability options for you to consider:

• Hot-spare drives

When a drive failure occurs, FlashDisk RAID systems contain parity data and thus can operate without downtime during a rebuild to a hot-spare drive; however, the array is vulnerable to a second failure during this time. With a hot-spare drive, the system will automatically rebuild the data from the failed drive to the hot-spare using the parity information, even if you are not there. You have the option to use multiple hot spare drives - in case you feel you need to protect against two or more drives failing before you arrive. Although this does not prevent a lost array if a second drive failure occurs before the successful rebuild of the first failed drive.

• Dual controllers

You can protect against controller failure with dual controllers operating in an active/active or active/passive configuration with mirrored cache to protect against data loss. As an alternative to relying on failover controllers, RAID 15 arrays offer field proven technology that sustains at least three simultaneous disk drive failures. RAID 15 also delivers high speed with dual independent controllers that maintain separate, yet redundant, write caches.

• Drives per channel

FlashDisk typically offer two to three disk drive channels, containing four drives per channel. If a system restart is an unacceptable delay, then RAID 15 is strongly recommended, especially in 24x7 operations.

• Drives per controller

Winchester Systems FlashDisk RAID arrays support up to twelve drives per controller. Speed is increased by striping data across multiple spindles. The performance threshold of twelve drives per controller has been determined through careful analysis and testing. A very high performance, high reliability configuration uses any where from four to eight channels where two to three of those channels are used as drive channels. The remaining channels are used for host connections.

• RAID 15

This special RAID mode is available on host servers that support mirroring of two RAID 5 arrays. It is the ultimate in high availability. RAID 15 supports the failure of every type of component simultaneously. You can lose at least three drives, plus fans, power supplies, channels, a host bus and a controller and still the server will run transparently!

• Redundant AC Power

RAID systems with multiple AC power cords can be plugged into primary power and secondary power sources. One power source may be an emergency generator in critical applications. A UPS is required when using write cache to protect against loss of data. An on-board battery backup to further protect the cache is an added precaution. Winchester Systems FlashDisk cache is generally 64 to 128 MB and can be flushed to disk within ten to twenty seconds in a typical system.

• Redundant power supplies and fans

Redundant components ensure that the array keeps running despite a failure of a power supply or fan.

• Hot swap components

Components that are hot swappable are mounted or attached in a very easily removable manner. They can be replaced while the system is operating under power, transparently to the users. This is important in order to avoid even a few minutes downtime for shutdown, replacement and restart of system if the time to perform it is prohibitive.

• Multihosted Servers

This configuration allows two or more servers to access a FlashDisk RAID array. Typically, the servers access separate data and can even run different operating systems.

• Clustered Servers

This is similar to the standby server but all clustered servers are in constant use, eliminating a wasted server. All servers have universal access to the data. Digital UNIX and OpenVMS, Windows NT, and Solaris provide the ability to support clusters. In some cases, additional software must be purchased to perform the clustering. For example, in the case of Solaris, Veritas Cluster Server software is necessary.

• Enclosures

It is easy to install and maintain arrays in enclosures with all disk bays initially full so that field upgrades are not necessary. At a minimum, make sure the enclosure comes with receivers for all removable disk trays.

contact us | site map | privacy
copyright © 2007 by Winchester Systems