On Thu, 26 Apr 2007, Florin Iucha wrote: > Nope, it was a USB 2.0 add-in card, and a Gigabit NIC. USB just sucks > for storage, especially when you layer it on top of IDE... On Thu, 26 Apr 2007, Nate Carlson wrote: > I find that firewire's much better - lower cpu utilization and such. On Thu, 26 Apr 2007, Chuck Cole wrote: > For those who think USB might be fast, it's worth pointing out that USB > is serial and has a protocol overhead as well. IDE has transaction > overhead also (quite visible when buffer depth is exceeded). The > fastest IDE is parallel and 3x faster than USB 2.0, ignoring protocols. > The oldest and slowest IDE is roughly the same as USB 2.0, however. > Ignoring overheads, USB 2.0 has a max of 480Mhz while SATA is 1.5GHz. > The real difference will have USB much slower than 1/3 the speed of > SATA. For a conversion to work at near the maximum speed, the > interfaces must be appreciably faster so the overhead is masked. > > USB is plenty fast enough for keyboards, mice, printers and modems, but > not much else. It's very convenient, commonly found, robust, and > inexpensive, so "good enough" for many uses. I didn't test it on a Linux box, but I bought a Western Digital MyBook Premium II 1TB external HDD last week. (It seemed nice at first but then it reported a corrupt drive.) I did some testing before it failed and compared performance of USB 2.0 with Firewire 400 on an XP x64 box, writing to the external HDD only: I wrote 8.734 billion bytes (1,757 files in 128 folders) via Firewire in 9 minutes 46 seconds which is 119 Mbps. It took about 2 seconds to delete them. Then I repeated the same file transfer via USB 2.0 and it took 9 minutes 4 seconds which is 128 Mbps. At first I thought this meant I should stick with USB because it happens to be easier to reach the USB port on my PC (and it was trivially faster than Firewire on that box), but now I think I should probably use Firewire, assuming you all are correct about the CPU overhead problem. That would be important. Mike