Test Setup
| Hardware | |
|---|---|
| Processor | 2x Intel Xeon, 3.6 GHz1 MB L2 Cache (Nocona) |
| Motherboard | Asus NCL-DSIntel E7520 ChipsetBIOS 1005 |
| Memory | 2x 512 MB DDR2-400 Corsair, ECC, RegisteredCL 3-3-3-10 |
| PATA/USB controller | Intel 82801EB (ICH5) |
| SCSI controller | Adaptec AIC-7902B Ultra320 |
| SATA controller | Silicon Image Sil3124 4-Port SATA-II-Controller |
| Graphics card | ATI RageXL , 8 MB |
| Network | Broadcom BMC5721 PCIe Gigabit Ethernet Controller |
| OS | Windows Server 2003 StandardService Pack 1 |
| Benchmarks & Measurements | |
| Performance measurements | c’t h2benchw 3.6Robocopy 1.96 |
| I/O Performance | IOMeter 2003.05.10Fileserver BenchmarkWebserver BenchmarkDatabase BenchmarkWorkstation BenchmarkThroughput Benchmark |
| Drivers & Settings | |
| Graphics driver | Windows Default Driver |
| IDE driver | Intel INF Drivers 7.0.0.1025 |
| DirectX version | 9.0c |
| Resolution | 1024x768, 32 Bit, 85 Hz |
New Benchmark : Simultaneous Reads And Writes !
To understand what this benchmark portends, consider this everyday occurrence : If you read a compressed Zip or Rar archive, its contents must first be extracted from that archive and decompressed to be read. Many compression programs write uncompressed data directly to the target media ; others put that expanded data into a temporary system directory before copying it into whichever folder the user selects. Whenever one also requests data from the flash drive at the same time this process is underway, unpacked files are being written at the same time additional files are being read.
When a 100 MB e-mail file is being copied from the flash drive to a computer and you want to write files to the flash drive at the same time, the same scenario unfolds. There’s nothing unusual about this, either.
Tests conducted while reading with simultaneous writes underway, and vice-versa, have revealed major differences among various USB flash drives. For this reason, we’ve added the following scenario to our normal test procedures. We request access to one file while copying another at the same time ; one of these files is 174 MB in size, the other 656 MB. Read and write operations start from different hard disks, so that no bottlenecks impact these two sets of operations.
For our read-while-writing benchmark, we initiate the copy of the smaller file onto the USB flash drive being tested first, then we start a copy of the large file onto the USB drive. Shortly thereafter, we initiate a second copy operation : this time the small file is copied from the USB drive to a second hard disk, and that transfer time is measured.
For the write-while-reading test, these operations are reversed : the big file resides on the USB flash drive being tested and the file is copied to a test hard disk. From a second hard disk, we then initiate a write of the smaller file to the test USB drive and that transfer time is measured.
To perform these tests, we use a copy program known as Robocopy.
