Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors Home PC Hardware from Made-in-China.com Search: Enter your search terms   Web www.lostcircuits.com Submit search form	.
Prices:

LOSTCIRCUITS		SHORTCUTS: The Challenge Fujitsus and Promises Assembling the RAID Building the System Test configuration HDTach vs. WB99 ATTO vs. Sandra Back to HDTach, WB99 Network File Copy Conclusions Fujitsu Drives Comments?
Internal 4-Way SATA RAID For The XPC     A Concept Study
(Review by MS, April 20, 2003)

One approach that will not give any accurate results is to copy files from one partition of the same drive to another. The reasons are simple, first, there will be bus contention, second, the drive itself has to constantly move the heads from one partition to the other, meaning that mechanical latencies as seek and rotational delay will become the dominating factors.

We took the easy way out and just used network transfer of files which does not exactly qualify as file copy but does fit into the targeted role of the setup. What we did was to copy Unreal Tournament (361 MB or 384,000 KB) via network from the SB52G2 to the Canterwood setup and back.

Surprisingly, or maybe not, copying the UT2003 folder from one system to another gave two different values, depending on the direction of the transfer. We hand-stopped the file transfers and calculated the averages of 3 transfers (with reboots in between each run). The numbers are rounded to the next full MB/s.

Regardless of which way one may look at this, the numbers are extremely impressive, that is, we are getting network transfer rates in excess of what most drives can actually handle in a peer to peer copy, at least as long as the SB52G2 is the source and the Canterwood is the target setup. Going the other way, we are still looking at some 25 MB/sec average transfers, which is in the order of drive-to-drive transfers of any early UATA-100 HDDs. There are several possible reasons for the difference in transfer rate, the different cache sizes, the different Gigabit Ethernet implementations or any other system-specific issues, and we cannot really pinpoint the source.

Conclusion

We set out to stick a 4 way RAID into a Shuttle XPC as proof of concept and we ended up with what could be described as the "Little Box From Hell" in an all-positive sense. Even the "limited" performance of these small drives is already pushing the envelope of the PCI bus, at least in a SATA 4 disk RAID 0 configuration. We had zero problems with the setup with respect to stability and reliability or else with going through all different kinds of configuration, once we received a production BIOS and drivers.

Keep in mind that what we were looking at were still first-release versions of BIOS, drivers and, not to forget, the drives themselves. The nex generation of these drives is already on the horizon, running 5400 rpm and an incredible 16 MB disk cache without changing platter area density or configuration. Based on these numbers, we can predict that the OD sequential transfers will be 30 MB/s which would propel a RAID 01 to 60 MB/s sequential R/W performance and a 4 disk RAID 0 beyond the capabilities of the PCI bus. On the other hand, the 16 MB cache, for a combined 64 MB in a four disk RAID 0 will probably allow Business Disk Winmarks in the order of .... at least 50 - 60,000 if not more. I can't wait to get my hands on those ....

Any further questions?

next page:    => More =>