How to select a RAID disk array
|"Most vendors will provide you with an implementation of RAID in which one extra disk is used for parity information to recreate lost data in the event of a single disk failure. "|
Shopping for a RAID disk array can be simple, once you know what you really need and understand what the technology can offer you. RAID disk arrays provide data storage with a high degree of operational availability and performance, depending on the feature set available from various vendors. All RAID disk arrays will provide you with some level of storage, reliability, performance, and service for different prices. Your job is to determine which is the best fit for you. In this article, I'll walk you through the decision process for choosing a RAID disk array.
All RAID disk arrays provide storage, but so does any old bunch of disks. Clearly, anyone can pile boxes on boxes to meet your total storage requirements. However, you need to consider these questions as you shop: How well does the storage scale from single units to racks of storage? Do all the disks use a limited number of host interfaces? Do all the disks use a very small number of controllers? These are sources of potential data bottlenecks. Does the vendor offer a rack cabinet? Who provides uninterruptible power supplies (UPSs), and are they adequate? Does the system require software drivers that must be updated constantly, or is the product a truly plug-and-play open system? Is the product scalable as your needs grow? Can it be moved to another platform and operating system easily and without added expense?
Most vendors will provide you with an implementation of RAID in which one extra disk is used for parity information to recreate lost data in the event of a single disk failure. You should avoid any vendor that does not meet this minimum criterion. Given that most vendors will offer similar capabilities on this most basic issue, you should ask more detailed questions: How long does the controller take to rebuild the data? Does the controller permit an automatic hot-spare replacement? How easy is it to monitor and control the status of the array? Do you need to be at the system to monitor and control it, or can you operate remotely--and if so, how?
RAID modes of operation
|RAID levels RAID levels 0 through 6 were defined in the original University of California Berkeley RAID project. RAID 2, 4, and 6 are rarely seen in commercial products. RAID 0 is merely disk striping, which has some performance advantages but stores no parity information and thus does not offer true RAID data protection. RAID 1 offers complete duplication of data, and this 100 percent data redundancy provides the best protection--but it is much too expensive for most applications. RAID 3 and RAID 5 each use one extra disk to store parity information needed to recreate data in the event of a single disk failure. RAID 3 uses a dedicated parity disk and is typically faster for throughput-oriented applications, such as file transfer and other sequential applications. RAID 5 distributes the parity information across all disks in the array and is typically faster for transaction processing and other random access applications. These results are relevant mostly in arrays that have little or no controller cache memory. In products with significant cache memory (64MB or more) on-board the controller, performance will be higher in all cases due to the distinctly higher abilities of the controller; these products will perform in a vastly superior manner regardless of RAID mode.|
Much has been written on the various RAID "levels." However, the word levels is a misnomer--it would be better to refer to RAID modes. Each RAID mode is just a different (not necessarily better) means of operation. (For more information, see the sidebar "RAID levels.")
You should insist that your RAID vendor support standard RAID levels including 1, 3, and 5. If the vendor has created additional RAID levels that are non-standard (that is, outside the range of RAID 0 through 6), be careful that you aren't buying into a proprietary architecture. However, certain RAID levels are merely combinations of two other RAID levels. For example, in RAID 1+0 (also called RAID 10), multiple RAID 1 pairs are striped for faster access; and in RAID 15, two RAID 5 arrays are mirrored for added reliability. These combinations offer advantages over single RAID modes and are perfectly acceptable.
Hardware fault tolerance
RAID vendors vary greatly in the degree of fault tolerance they provide. To maximize system availability, you need redundancy in other system components that are most likely to fail, including power supplies and fans. (Disk controllers, unlike all the other mechanical components, are completely electronic and thus are the most reliable components. You can optionally look for redundancy here, too.) To the extent possible, these components should be hot-swappable so that they can be replaced while the system is running, further increasing total system availability.
You should also make sure that each individual enclosure has two AC power cords. Install one into a dedicated UPS and one into a room outlet (preferably a protected outlet) or second UPS. Any RAID disk array with cache memory on the RAID controller should also have a battery backup module for added protection.
Quality construction is an often-overlooked parameter in today's world of sleek-looking enclosure design. Don't be fooled by the smooth, rounded corners of plastic disk enclosures: Only all-metal disk carriers offer the thermal conductivity needed for proper heat dissipation for today's high-speed, higher-capacity disk drives, which often run hot and require proper operating environments. Proper cooling is critical to achieve long and reliable disk-drive life, and metal carriers simply do this better.
Metal carriers also shield the disk drives' high-speed signals from stray radio frequency interference (RFI) that may occur in an office or computer room environment. For 18GB drives and high-speed 10,000rpm drives, it is also important to minimize drive vibration to avoid excessive disk errors and time-consuming retries. A quality vendor will have a new drive-mounting scheme to ensure reliable operation of 18GB and 10,000rpm disk drives--the vendor won't just put these more sensitive drives into the same old plastic disk carriers that handled their 1, 2, 4, and 9GB drives.
Performance is a critical parameter for every server. After you spend all your money on a RAID disk array, is it going to give you the performance you need? Today, at competitive prices, you can get over 9,000 disk read/write operations per second for transaction-oriented applications and up to 35MBbs actual sustained throughput for data transfer operations. If a vendor is not providing anything near these specs, all your applications will run needlessly slowly. Today, computers process data in near zero time; thus, all server applications run in proportion to the speed of the slowest device, which is the mechanical disk drive.
Ask vendors for the performance specs of their RAID disk arrays, and compare them. Don't accept generalizations. If they are clueless or do not publish their specs, you can be assured that they do not measure up.
You need to ask vendors a number of questions regarding service. Can your people service the unit, or must you rely only on outside service providers? How easy is it to replace disk drives, power supplies, fans, and controllers? Can any moderately skilled technician perform these component replacements? How quickly can the components be replaced? Are they hot-swap replacements--can components be replaced while the system is still delivering data to users who are unaware of the problem? Does the vendor have on-site service available? Do it have an 800 hotline staffed by people who actually know the equipment? Is the service available on a 24x7 basis? Can you or a highly trained factory engineer dial in to your system from a remote site? Can your system automatically alert you via pager in the event of a warning or error condition?
|"When you're comparing systems, vendors who are willing to guarantee performance in writing are more credible. "|
The vendor you select is as important as the product! Is the company merely a distributor of the product that may not know much about it, or does the product come factory direct from the people who know the equipment? Are you considering buying the storage from the server manufacturer because doing so is most convenient, or are you truly looking at your needs and choosing the right RAID disk array? Is the vendor completely committed to RAID disk array technology, or does it sell many other products that dilute its interest and expertise? Is it important to you to have the comfort that comes with a name brand ("no one ever got fired for buying IBM"), or do you just want the best product for your needs from a company that can stand behind the product?
How long has the company been in the business? Does it enable you to be self-sufficient in a crisis or leave you completely dependent? Does it use industry-standard components? Does it offer systems designed to open-systems standards, or has it managed to include proprietary components that lock you in to its architecture? Does it use the industry's best disk drives or less expensive models with correspondingly lower quality and reliability?
Is the price fair, compared to other vendors' offerings, or it too high or too low? Do the company's representatives in sales, sales support, and technical support seem to have the expertise needed to support the product? Do they exhibit genuine interest in providing solutions to your needs? Are they enthusiastic about their products and committed to showing you how much they can be of service--or is their zeal devoted to separating you from your budget as quickly and efficiently as possible?
You should ask the vendor for references from other customers. Ask the company's customers how well they were treated before and after the sale. Did the company meet its promises? Did the product live up to its claims? Were the performance, reliability, and service delivered? How smoothly did the installation process go?
You can be sure that a company will treat you much the same as it treated others. Ask people if they are happy with their purchase and if they would buy from the company again.
When you're comparing systems, vendors who are willing to guarantee performance in writing are more credible. If a vendor claims to offer a high-performance system that runs fast, see if its will back up its claim. A vendor that believes its claim will gladly guarantee the results you want, because it has confidence (based upon prior experience with other customers) that you will get the performance you're paying for.
Try to be as specific as possible with respect to your most important application. For example, challenge the vendor to cut your month-end report times by 33 percent, to serve twice as many Web pages from your server, or to cut lengthy database inquiries from five seconds to two seconds. That way, you can test the RAID disk array immediately upon receipt and know right away if you received the advertised benefit. Such a guarantee also builds a case to justify the investment to management in the first place, and communicates clearly to the vendor what you expect in return for your hard-earned cash. It also filters out vendors who know they can't really deliver. By being specific and getting a guarantee, you simplify the purchasing process and the management justification process, and you increase your odds of success while minimizing risk and conserving your valuable time.