System setup and notes
The i-RAM fancies itself as a fast alternative to a hard drive, so we thought we'd put it up against a pretty fast RAID array.
| Motherboard | ASUS A8N SLI Deluxe |
|---|---|
| CPU | AMD Opteron 146 (2.0GHz) |
| RAM | 2x512MiB Corsair Value Select DDR400 - 2.5-3-3-8 |
| Disk controller | nForce 4 SATA controller |
| Relevant drivers | Windows default for disks, 5.10.2600.552 for NVRAID |
| Operating System | Windows XP Professional x64 Edition, SP1 |
| Disks | Gigabyte i-RAM (2x 1GiB OCZ PC3200 installed) 2x 250GB Seagate ST3250823AS in SATA RAID-0 2x 250GB Seagate ST3250823AS in SATA RAID-1 |
Testing software
IOMeter 2004.07.30
IOZone 3.259
HD Tach RW 3.0.1.0
Notes
The test system had four 250GB disks using up all four of the NVIDIA motherboard's SATA ports. In light of that, we initially made the mistake of hanging the i-RAM off the Silicon Image SATA controller present on the board. This controller is sat on the PCI bus, which provides a maximum of 133MB/s, and that bandwidth is shared between all PCI devices. Seeing as SATA can reach 150MB/s, and we were testing a RAM-drive, we took one of the NVRAID storage arrays offline and hung the i-RAM off that instead. The NVIDIA SATA controller sits happily on the northbridge and, oh, what a difference it makes. So, if you can, use an i-RAM on a controller that isn't held back by the PCI bus.
We ran the default (quick) (8MB zone) HD Tach tests, but they yielded similar results to our IOMeter customised tests, so we only decided to present its burst rate figures. With IOMeter, we used the following access specifications:
| Option/Test | Configuration |
|---|---|
| Outstanding I/Os | 10 |
| Individual test run time | 30 seconds |
| Read test access spec | 1MB transfers 100% sequential 100% read |
| Write test access spec | 1MB transfers 100% sequential 100% write |
| General usage access spec | 64KB transfers 50% sequential, 50% random 33% write, 67% read |
The 'general usage' spec. shouldn't be taken too literally. It's basically an access spec. that throws a lot of read/writes at the disk, rather than sequential transfer rates. It doesn't represent any particular scenario, but it does demonstrate how a disk will perform when it's not just being read/written sequentially.
IOZone is a filesystem test and so can present lots of varied information, not all disk-related. However, issues such as response time can greatly affect the overall read and write rates reported by IOZone. We tested the i-RAM using file sizes of 4KB through to 128MB, with record (i.e. transfer) sizes of between 4KB and 16MB.
