The following system was used for all benchmarking. It reflects a typical SME environment system.

• Dual AMD Athlon MP 2400+
• Asus A7M266-D Motherboard BIOS 1.10
• 2048 MB PC2700 ECC Registered DDR RAM - four 512 MB DIMMs
• Maxtor DiamondMax Plus 9 80 GB boot drive - Primary Master
• ATI Xpert 98 8 MB AGP with OS drivers
• Sanyo 52X ATA33 ATAPI CD-ROM - Secondary Master
• 3Com 3C905C-TX 10/100 PCI NIC with OS drivers
• Antec SLK3700AMB case with Antec 430 watt power supply and two 120mm case fans
• Maxtor DiamondMax Plus 9 80 GB - four drives as Parallel ATA RAID array
• Maxtor DiamondMax Plus 9 120 GB - four drives as Serial ATA RAID array
• Microsoft Windows 2000 Server with Service Pack 4 - minimal installation

Testing is done with the following benchmarks:

• VeriTest WinBench 99 2.0
• TestaCD Labs HD Tach 2.61
• IO Meter 2003.02.15 using Storage Review’s testing methodology
• ATTO Utilities Version 1.63 Windows Disk Benchmark 2.02
• Futuremark PCMark2002

This system best fit the concept of a SME, or small to medium enterprise, server. The focus of our ATA RAID testing is on the environment, not onboard RAID or two disk RAID 0 or 1. Four disk RAID 0+1 or 10, 5 are the configurations looked at and benchmarked.

All adapters are benchmarked using a 66 MHz 64 bit PCI slot on the A7M266-D. Settings are left at the default parameters, cluster size at 64 kilobytes.

Why these benchmarks? I’ve done a lot of reading on the net, of other site’s RAID and hard drive reviews. These tests should provide a solid picture of what the ATA RAID cards can do. WinBench and HD Tach are industry standards. IO Meter has had a falling from grace lately, but look at what Storage Review had to say:

IOMeter unfortunately lacks one key setting: "Restrict xx% of access to within yyyyy sectors." Such a parameter would permit, say, 50% of accesses to occur within 8 megabytes of the last request, effectively simulating locality. As it stands, however, IOMeter is tilted towards accesses that span the entire drive without regards to locality- a pattern that much more closely simulates multi-user server environments rather than a single-user desktop/workstation.

While this limits usability for single drive tests, it’s perfect for our testing methodology. The ATTO benchmark gets a lot of talk amongst users in the discussion boards, so it’s included. Futuremark is a big name in graphics benchmarking, but I liked their simple disk benchmarks as part of the PCMark2002 suite.

IOMeter and HDTach require unformatted drives to run. The rest require a formatted partition. For these benchmarks I formatted using NTFS and it’s default cluster size on a basic disk. There has been much speculation regarding increased performance using dynamic disk in Windows 2000, but that wasn’t evident in these benchmarks.

Look for Platform 2004 to have Windows Server 2003 and a few new benchmarks, as well as strict focus on Serial ATA drives.