Re: [Libusb-win32-devel] How Fast?

[Brad K. <kin...@ho...>]

Dan,

Thanks for the prompt reply. Is there an example on the website, or 
somewhere else, of the double buffering and async call technique?

The target hardware we are trying to communicate with sounds similar to your 
setup where we have an FPGA consuming the data via DMA, except it is not 
using NetChip hardware. I don't know what USB hardware it is using as it is 
being custom built by a subcontractor. It is good to know that such high 
rates are possible, since that is what I'm trying to prove to them.

Did you use any special host controller hardware on the PC in order to get 
the 260Mbps rate? I read something in the list archives that some 
controllers are slower than others.

Thanks.
Brad


>From: "Dan Ellis" <Dan...@di...>
>Reply-To: lib...@li...
>To: <lib...@li...>
>Subject: Re: [Libusb-win32-devel] How Fast?
>Date: Fri, 17 Nov 2006 13:42:57 -0000
>
>Brad Kingston wrote:
> > What is the practical sustained speed limit I should expect when
> > using the usb_bulk_read and usb_bulk_write calls? I'm seeing a limit
> > of about 20MB/sec when talking to a PXI NetChip peripheral card and
> > another custom board. I'm not sure if its the libusb side or the
> > peripheral side that is limiting me to 20MB/sec.
> >
> > In a tight loop that sends 1GB of data, I send 150MB in each
> > usb_bulk_write call in order to get up to the 20MB/sec speed I'm
> > seeing. As I drop the size down for each write, the sustained rate
> > decreases also.
>
>We've managed to get sustained 260Mbps (32.5MB/s) doing bulk writes to a
>NetChip NET2272 with it operating in 16 bit DMA mode (and using libusb
>with 2 buffers and async transfers). We had an FPGA consuming the data
>from the DMA. I don't think the NetChip could achieve that rate on the
>PCI card.
>
>We never tried bulk reads.
>
>The bus won't accommodate more traffic than that - there was no NAKing
>going on, the gaps between the transactions looked on the analyzer to be
>about long enough for the data to be read over PCI ready for the next
>transaction, so I think PCI is the limiting factor. It's possible that a
>read could get more data bandwidth out of the bus.
>
>In 8 bit mode this drops to 180Mbps.
>
> >From a software perspective, it's quite important to use double
>buffering and the async calls so that you keep the controller busy with
>work to do. On Linux which doesn't have an aynchronous interface to
>libusb (yet!) I'm struggling to get more than 140Mbps.
>
>--
>Dan Ellis
>Senior Development Engineer
>DisplayLink (UK) Ltd.
>
>
>-------------------------------------------------------------------------
>Take Surveys. Earn Cash. Influence the Future of IT
>Join SourceForge.net's Techsay panel and you'll get the chance to share 
>your
>opinions on IT & business topics through brief surveys - and earn cash
>http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
>_______________________________________________
>Libusb-win32-devel mailing list
>Lib...@li...
>https://lists.sourceforge.net/lists/listinfo/libusb-win32-devel

_________________________________________________________________
MSN Shopping has everything on your holiday list. Get expert picks by style, 
age, and price. Try it! 
http://shopping.msn.com/content/shp/?ctId=8000,ptnrid=176,ptnrdata=200601&tcode=wlmtagline