For the IO gurus

[PAPutzback]

After seeing the responses and of the few there were in the general forum I figured my question might be to technical.

When hooking up external hard drives via usb or firewire what are the requirements of the CPU to just share data out.

How does the CPU handle the IO when it will actually not be using the data for its own purpose (Watching a movie or loading game data into memory).

If this is to low tech for this forum delete it and I'll understand. but the reposnes to file server questions in the general forum had no sensible logic behind there replies.

I have a 500 P3 with a firewire card hooked up to a bridgeboard with two drives on the bridge. Now if laod the enclosure up with 8 drives across two bridgeboards connected to the pci board and data is being pulled from 2-4 drives how much does the CPU play in the transfer.

Thanks.

 

[Bassyhead]

Firewire utilizes DMA which means the CPU has little to do with the actual transfer of data. Data between Firewire devices and memory are exchaged directly by a DMA controller, so CPU utilization will be very low.

 

[JHutch]

Bassyhead is right about the DMA.

That being said, with a P3/500, you are going to be limited by the bandwidth of the firecard, not the CPU. Everything is funneling back through a single firewire channel, which means that all the drives together can output far greater bandwidth than the FW card can handle.

In my case, I have a 400MB/sec firewire card pulling from a chassis containing 8 drives (for a total of 1.1 TB and yes I am proud of that! ). The CPU (P3/450) barely breaks a sweat no matter how hard I hit the drives, because of DMA access.

And in my case, since that computer is merely a network file server to the rest of the house, it is limited by the 100M ethernet line anyway. (But even internal transfers inside the computer, which max out the firewire, can't max out the CPU)

JHutch

 

[uOpt]

I never benchmarked this, I am curious, too.

But it is much more complicated than just DMA or not.

In particular:
- there might be block size restructions that cause the USB/FW connected drive to generate more interrupts
- there might be braindeadness in the USB/FW controller chips that force the drive to copy data around at interrupt time, similar to the Realtek 8139 ethernet cards
- I am pretty sure most USB/FW controller chips these days sit on the 32bit/33MHz PCI bus, which is insufficient for today

 

[TerryMathews]

Believe it or not, one of the most potent platforms for this application (Firewire to Ethernet) would be the i875, with it's CSA GigaBit lan (basically dedicated PCI-X bus) and on SouthBridge IEEE1394.

Most chipsets have moved away from IEEE1394, so performance has technically degraded if you needed IEEE1394 bandwidth.