While it is more or less clear with graphics cards, many users want to know what power supply they need in order to build a RAID array. To answer this question I took the same configuration as in the previous section and added three Western Digital Raptor WD740GD hard disk drives. These are not very new and not very economical of HDDs. They were connected to the chipset’s controller and united into a RAID0.
- Processor: AMD Athlon 64 X2 5000+ (2.60GHz)
- CPU cooler: TITAN DC-K8M925B/R
- GlacialTech SilentBlade II GT9225-HDLA1
- ASUS M3A78 (AMD 770 chipset)
- System memory: 2x1GB Samsung (PC6400, 800MHz, CL6)
- Hard disk drive: 250GB Seagate Barracuda 7200.10 ST3250410AS
- Graphics card: 512MB Sapphire Radeon HD 4650
- Optical drive: DVD±RW Optiarc AD-7201S
- System case: IN-WIN EAR-003 (400W)
- Hard driives: 3 x 74GB Western Digital Raptor WD740GD
I installed Microsoft Windows Vista Home Premium SP1 (32-bit) and necessary drivers on the PC.
The disk subsystem was loaded by means of a special program of our own writing. We had written it a few months previously for quite different purposes.
FC-Verify can create and read a specific file set in two independent threads. Thus, at any given moment, there is one read thread and one write thread, which is quite a serious load on the disk subsystem. The files are accessed using Windows API; file caching is disabled; the data block size is 64KB. Besides, the program verifies if the files are read and written correctly, but that’s irrelevant for my purpose. There is a 10-second pause between writing and reading in each thread. The files are deleted after each write-read cycle, and the cycle is repeated from the beginning.
I made the program process a thousand 256KB files in one thread and a hundred 10MB files in the other thread as you can see in the screenshot. The computer’s power consumption was being measured continuously through a few write-read cycles.
But first goes the system boot-up stage with one disk only: the system disk is working while the Raptors are as yet turned off. The diagram does not show anything exceptional besides the longer time it took to enable the CPU’s power-saving technologies. This was because the chipset’s RAID controller took a long thought over the identified system disk and the turned-off array.
This is the same load but the RAID0 array out of three Raptor WD740GD disks is turned on. The most interesting thing is the tall peak at the beginning of the graph that corresponds to the HDDs’ spinning up their platters. The total consumption from the +12V rail (CPU, mainboard and drives) is over 11A then.
It is the +5V rail that is loaded the most here. Clearly, this is due to the HDDs’ electronics as well as to the chipset’s South Bridge the integrated RAID controller resides in.
It is also interesting that the +5V rail is under the biggest load with the RAID array, too. This can be explained, though. A movement of a disk’s head provokes a short spike of current on the +12V rail, but these spikes do not affect the resulting graph much because the array’s three HDDs do not move their heads in sync. Anyway, the diagram shows this in a clearer way.
Somewhat surprisingly, the hardest moment for a file server is when the spindles of all the HDDs are spinning up simultaneously. In the process of normal operation the HDDs’ electronics load the +5V rail considerably while the +12V consumption is rather low.
Thus, a typical 300W power supply is quite enough for powering a 3-disk RAID array with rather voracious HDDs. Such a PSU will easily start the system up and will ensure a threefold reserve of power at work.
I can also note that each fast HDD requires an additional 3.5A on the +12V rail at startup. For large disk arrays based on Raptor-like HDDs it would be good to have a smart RAID controller that can turn HDDs on one by one when the system is started up.