What to Look for When Selecting a Scalable RAID Array

Suppose you're looking for a robust and scalable RAID array solution to store your important enterprise data. You need to consider the fault tolerance of the RAID array, the RAID array manageability tools, support for dynamic growth of the RAID array, support for additional disk drives, and the I/O technology in place.

FAULT TOLERANCE OF THE RAID ARRAY
Regardless of how well you tune and size your RAID array, it can still fail to a point at which you can't access your data. The importance of your data directly affects the level of fault tolerance you apply to your RAID array solution. Improved fault tolerance is synonymous with higher cost. If you use a RAID level other than RAID 0, you've already decided that your data has some level of importance because you're providing disk-level redundancy in the form of RAID. Beyond this precaution, consider these desirable fault-tolerant features to further bolster your overall RAID array solution: redundant hot-swappable power supplies, redundant hot-swappable fans, and hot-swappable hard disks. These items are becoming more common, and you should consider them mandatory.

Some external RAID arrays, such as those from Data General and Symbios, provide embedded RAID array controllers. These units internally support multiple SCSI channels to each hard disk, so that if one disk array adapter channel fails, you can still access the disks in the array. For the ultimate in RAID array reliability, you can add multiple RAID array adapters in your server, where each provides a separate channel to your RAID array. Again, if one RAID array adapter fails, the server and subsequently your customers can still access the information on the array. You can also leverage these extra channels for improved performance of larger RAID arrays. Whenever possible, try to remove single points of failure.

RAID MANAGEABILITY TOOLS
Manageability tools are the most commonly overlooked component of a scalable and reliable RAID array system. How do you know whether a disk in your RAID 5 array fails or a power supply fails? Unfortunately, such failures can kill performance because a RAID array that is operating in fault mode (e.g., with a failed disk in an array) runs slowly. You need to ensure that your RAID array includes built-in manageability tools, and you need to learn how to use them. Typically, these tools let you configure and monitor your RAID array. These tools often include an array agent that runs as a service under Windows NT. The array agent closely monitors the performance of your RAID array solution. If a disk or another component in the array fails, the array agent records the failure in NT's event log. Also, you can configure array agents to send email, provide a visual alert, or even page you as needed (depending on the vendor's implementation). Diagnostic lights on the outside of the RAID array are cute, but who wants to sit in a computer room and watch some lights blink on and off or listen for audible alerts? Instead, you can easily automate the diagnostic monitoring and ensure that your RAID array is running flawlessly.

SUPPORT FOR DYNAMIC GROWTH OF THE RAID ARRAY
To configure a RAID array under NT a few years ago, you had to boot your NT system into DOS mode, run a utility to create the RAID array, and then format the array under NT with a file system. Later, if you needed to add an extra disk to the array for capacity or performance, you had to back up the data on the array, go into DOS mode, reconfigure the array, reformat the array, and restore the data. Fortunately, you don't have to endure this process today. When selecting your RAID array, ensure that it includes tools to let you dynamically add hot-swappable disks. This functionality allows for a truly scalable RAID array solution.

SUPPORT FOR ADDITIONAL DISKS
Now that you can dynamically add disks to your array, ensure that your RAID array supports additional disks. Wide SCSI-2 lets you connect up to 16 devices to one SCSI channel. Depending on your workload, you might consider configuring up to 10 disks in one large array and still not saturate the SCSI channel. You need to ensure that your RAID array enclosure will support the disks that you need in the future. Also, higher-density SCSI adapters now support one to four channels per adapter. This support lets you conserve those precious PCI expansion slots in your NT system. For even more flexibility when adding and configuring your RAID arrays, you might consider a RAID array system that leverages fibre channel technology. Both fibre channel- and SCSI-based RAID arrays must still wait on the physical limitations of the disks when obtaining data. However, fibre channel allows for more arrays per channel and greater distances between the server and its arrays.

I/O TECHNOLOGY IN PLACE: FOLLOW YOUR DATA
Regardless of whether you leverage fibre channel- or SCSI-based RAID arrays, you must closely monitor the data path between the RAID arrays and your server. Each disk can provide a certain amount of throughput. As you group more and more arrays on a particular channel, you need to ensure that the data path between your arrays and your server doesn't become saturated. To avoid saturation, select the RAID array technology that meets your needs. Fibre channel can run at speeds up to 100Mbps, and the latest SCSI standard can support speeds up to 80Mbps. Follow the data path from the disk to your NT system. You can take this precaution to ensure that you haven't configured too many RAID arrays and avoid overwhelming the data channel, the disk array controller in the RAID array, the RAID adapter in the NT system, or even the PCI bus that the RAID adapter attaches to.

NT doesn't provide any one counter in Performance Monitor to accomplish this task, but you can review Performance Monitor's LogicalDisk object's Disk Bytes/sec counter for each array connected on a specific channel, add these values together, and ensure they aren't exceeding 70 percent of the theoretical throughput your data path can support. If the total does exceed this threshold, consider adding a second data (SCSI or fibre channel) connection to your RAID array configuration. The converse of this concept is helpful also. If you have plenty of bandwidth available from your current data channel, you can use this data channel even further by adding more RAID arrays with confidence.