mirroring, splitting and hiding
Another collection of features, when used in conjunction with each one another, delivers extra data protection from system threats derived from malicious users, accidental deletion or viruses. These data protection features include N-way mirroring, array splitting and array hiding. N-way mirroring goes beyond simple two-way mirroring, allowing additional mirrors of a data set to be created; array splitting allows one of those mirrors to be removed from the active array; and array hiding makes an array invisible to users and the operating system, and accessible only to a privileged administrator.
When combined, these features allow a system administrator to create a secure backup of the active data array. To start, the admin creates a three-way mirror using N-way mirroring. Next, he removes one of the mirrors from the active array through array splitting. (Note that the active array retains the data integrity provided by a two-way mirror after one of the mirrors is removed.) Finally, he hides the split mirror so neither users nor the operating system can see the data. Threats can't attack what they can't see.
Should a disaster occur in the active array occur, the fix is simple and local. The admin deletes the data in the corrupted array, unhides the hidden array and transforms the active array using ORLM to include the good mirror. Please note that the data was valid as of the time it was split off, so any modifications since that time would not be reflected in the previously hidden array.
Complemented by hot swapping capabilities, drive roaming makes moving disks between systems much easier by eliminating the need to keep track of which drives were connected to which RAID controllers. With drive roaming, the system keeps track of this for the user. This is extremely useful for applications that require moving large amounts of data between systems quickly. In such applications, it is much faster to move drives than to copy the data over the network.
Controller spanning allows an array to span disks attached to multiple RAID controllers and, in doing so, allows the creation of very large arrays and provides high throughput rates. For example, if we have four RAID controllers and each controller has eight drives attached, controller spanning allows the creation of an array that spans all thirty-two disk drives. Since performance scales linearly, this allows an extremely high I/O transfer performance with thirty-two spindles in a single array.
The final enterprise-class RAID feature is distributed sparing, which creates a spare failover drive without the need to actually include an extra drive. Essentially, distributed sparing reserves enough disk space on all the included disks so that the sum of the reserved space is equal to the largest drive in the array. A key advantage of this approach versus global or dedicated sparing is that all of the drives are actively used, resulting in significantly better performance. This is different from the standard, nondistributed approach, which sets aside one drive for failover, allowing the possibility of a silent failure when that drive is not in use.
There is some serious power in the enterprise-class RAID features that have traditionally been available only to those with large IT budgets. Now, with the convergence of RAID and lower-priced storage technologies, this same power can be enjoyed by any small-to-medium sized organization with critical data.
For more information:
About the author: Randy Arnott is a RAID architect at Broadcom specializing in software for embedded, card-based, and host-based RAID solutions. He has over 25 years of experience as both a hardware and software engineer with established storage companies like DEC, EMC, and Adaptec, as well as with two successful startups, Data Kinesis and RAIDCore, recently acquired by Broadcom Corporation.