RAID has been a system implemented in high-end servers and workstations for some time. Until recently however, it was not viable for most organizations. Now that it is an option, learn the ins-and-outs to determine how your organization can benefit.
By Dan Mepham
Of course, a large part of making RAID affordable lies not only in reducing the cost of the controller itself, but also in making it compatible with less-expensive hard disks. SCSI RAID controllers have been available for quite some time, but most organizations cannot afford a single SCSI drive, let alone multiple drives. Thus, the next logical step in making RAID affordable would be an IDE RAID controller. With that in mind, last year Promise Technology introduced its FastTrak. Following its lead, Iwill also recently announced sIDE RAID66, an UltraDMA-66 IDE RAID controller.
Comprehensive reviews of both products can be found on HardwareCentral. This tutorial will primarily discuss the concepts of RAID.
RAID 0 Striping
Suppose you have a 100 MB file (i.e., the milkshake) to write to a 10 GB hard disk. That hard disk (the straw) can write just so fast and no faster, so the file will be written (like the milkshake being sucked up) in a given amount of time, as fast as that hard disk will allow. Now, suppose instead of one 10 GB disk, there are two 5 GB disks of identical speed. Instead of writing 100 MB to a single disk, 50 MB can go to each disk. Writing to those two 50 MB chunks simultaneously effectively halves the time required to write 100 MB. Continuing that, suppose you used four 2.5 GB hard drives, and wrote 25 MB to each. Now, the 100 MB file can be written in the time it takes to write 25 MB to one drive. This is known as RAID striping.
Likewise, disk read performance will improve. Two drives can read 50 MB each twice as fast as a single disk can read 100 MB.
Essentially, RAID 0 is used to improve performance by using multiple drives, writing to each in "stripes." For example, using a stripe size of 8 KB, when data is sent to the drives the RAID controller divides the data into 8 KB chunks, and writes every even chunk to one drive and every odd chunk to another, effectively doubling read/write performance.
When using a RAID 0 array, you should attempt to use identical hard disks. If one disk is much slower than the other, much of the performance increase will be negated, as the faster disk will be left waiting for the slower.
The size of a RAID 0 array is dependent on the size of the smallest disk in the array. The total size is equal to the size of the smallest disk in the array, multiplied by the number of disks. For example, for an array of three 1 GB disks and one 10 GB disk, the array size would be only 4 GB (smallest disk = 1 GB, x 4 disks = 4 GB), as the remaining 9 GB on the 10 GB disk cannot be accessed by itself. This is another reason why identical drives should always be used: to avoid loss of capacity.
RAID 1 Mirroring
The unfortunate, obvious downside to RAID 1 is the loss in capacity. As noted, two 5 GB disks in a RAID 1 array appear as a single 5 GB disk. Although the increased reliability of this method cannot be disputed, the halving of disk space translates into increased cost.
Again, when using RAID mirroring, the size of the disk is equal to the capacity of the smallest disk. Using a 1 GB and 10 GB disk in a RAID 1 array is unwise, as it will appear as only a 1 GB disk the remaining 9 GB on the 10 GB drive cannot be used. Given that, use only disks identical in size and performance in a RAID 1 array.
RAID 0/1 & Spanning
Be aware, however, that to use a RAID 0/1 array you must have at least four hard disks (two groups of two striped disks). More disks (e.g., 8 or 16) can of course be used, should the organization desire. Again, it is best to use identical disks, as not doing so will result in a loss of performance or disk capacity.
Spanning is another function available on some RAID controllers. Spanning is a somewhat minor feature, as it does not boost either performance or reliability, but just adds convenience. Basically, two drives, perhaps of different sizes and speeds, can be joined to form one large one. Again, this is a somewhat moot point, as some operating systems (e.g., Win NT) can do this on their own; however, its at least worth mentioning.
It should be noted that RAID 1 arrays do not differ greatly from single disks in terms of performance. In fact, RAID 1 typically incurs a tiny performance hit, but not enough to worry about. In any case, the main focus of RAID 1 is reliability, not speed. There are, however, no hard and fast numeric reliability benchmarks.
While RAID won't benefit everyone, it will some. Those who want a RAID platform but are put off by the cost of SCSI, IDE RAID may suit perfectly. Iwills sIDE RAID66 is an excellent contender, performance is top-notch, and quality and reliability are excellent. And at only $99 MSRP (likely cheaper street prices), it won't leave your wallet dry.
More details about Iwill's sIDE RAID66 can be found here on hardware central. Dan Mepham welcomes any questions, comments, or fireballs you may have.
This article was originally published on Friday Apr 7th 2000