Dakota81,
There are a couple of things to point out about the article that you referred to. The first point is that it was published in July of 2004. Obviously there have been significant improvements in the past 17-18 months. Keep in mind that the RAID controller being tested was one of the first onboard controllers from Intel. Since then there have been a plethora of newer and better performing onboard controllers, some of which are actually integrated into the Southbridge chip (like nVidia's).
Secondly, I take issue with statement that most controllers are relying on software drivers for the RAID capability. Since Windows NT, the OS itself has had the ability to do software RAID, so there is little point to supposedly adding a chip that allows you to do software RAID. Furthermore, on the onboard RAID controllers that I have seen, the RAID configuration is handled in the controller BIOS, before the OS is even installed. If such is the case, then the RAID volume is already created BEFORE any Windows RAID drivers are installed, and Windows should see the RAID set as a single volume.
Back to the article, they demonstrated that in mid-2004 you could get a 20% performance improvement in pure IO tasks by RAIDing a pair of Raptor drives. Unfortunately, they only show the Winstone scores and not the actual throughput numbers. I would be very interested to know what those actual throughput numbers were.
Why? Because there is only a finite amount of bandwidth available to the PCI bus that the disk controller resides on, and that bandwidth is shared with other PCI devices (NIC, sound chip, etc). If a single disk drive is close to that limit, then doubling the theoretical performance will not double the actual performance, because you will hit the limit for the controller's interface to the PC. But on top of that, you're not just sending data to the disk controller. You have to get it to the CPU, which means that you have to cross the link from the Southbridge to the Northbridge, and then across the FSB to the CPU (both paths are also shared with even more devices). So in more modern PCs (those with higher FSB speeds or utilizing hypertransport instead of an FSB) this limit will be higher and you would be likely to see higher performance.
Anyway, because they only tested RAID with the fastest hard drives available (the Raptor II 74 GB drive) they guaranteed that they would be using a device that is already close to the bus limit. But if you went with a more mainstream 7200 RPM SATA drive, you would probably see larger performance increases. This is an important distinction. RAID stands for Redundant Array of Inexpensive Disks. It's all about getting the most performance without having to buy specialized or expensive hardware. So if you had the choice, would you spend $330 on a pair of 74 GB Raptor II drives in a RAID 0 array, knowing that you would max out the interface? Or would you rather max out that interface with a pair of 160 GB 7200 RPM drives that you could get for $180 or less? That would be twice the storage, the same speed, and half the cost.
Also, the overwhelming majority of tests were performed with general office applications that were not disk intensive. In fact, those benchmarks are generally used to test the performance of the CPU and memory subsystem. This GUARANTEES that the results will not show significant improvements in disk performance. That's fine if all you're doing is typing email and word documents, but there are quite a few users out there who do work that is at times disk intensive, and you would definitely see greater improvements if you benchmarked on those applications.
The other point to keep in mind is that all of their tests were performed in a single-tasking environment. Most people run more than one application at once. SMP systems and multi-core CPUs are becoming more common, which means that people will be performing more tasks simultaneously. If multiple threads both need to access the disk at the same time, a higher performing disk subsystem will improve performance. This is why they mentioned at the end of the article that they still considered RAID appropriate for servers (servers are usually doing several things at once).
So keep all of this in mind when evaluating desktop RAID solutions.
