Re: [Discuss-gnuradio] Gigabit ethernet

[Ralph Hyre]

I think firewire2/IEEE1394b should also be looked at, I realize there
are potential issues with chipsets, but it would be compelling to
integrate with existing firewire capture devices like disk, displays, and
cameras.  The low jitter and daisy chaining capability is compelling if
you want to 'stack' receivers.

Excerpt from http://www.tomshardware.com/howto/20040823/firewire-06.html:

>The new high-speed variation of IEEE1394 is popularly called FireWire-800 or
>FireWire-b. This is not quite correct as IEEE 1394b has not yet been accorded a
>trivial appellation. Officially, the standard is currently simply
referred to as IEEE
>1394b. Besides new connectors - from the front these look like a
larger version of
>the i.Link plug (about three times as big) - new drivers ensure a
performance leap.
>Maximum gross throughput rates in an ideal environment are around 100 MByte/s
>(800 MBit/s). In practice, average throughput in measurements of connections
> from the PC to external devices such as hard drives or DVD drives usually only
> reaches half of that (50 MByte/s) and direct connections between two computers
> only a quarter (25 MByte/s) of the maximum data rate of 800 MBit/s = 100
> MByte/s. Despite Ethernet's collision management (CSMA/CD), Gigabit
>-Ethernet networks attain average throughput rates of 70 to 80 MByte/s
>(effective). Collisions occur when two data packets are sent simultaneously.  

>As a serial transmission technology, FireWire avoids collisions
through intelligent
> time management - unlike Ethernet - so that jitters are in the pico-second 
> time
> range, in other words over three decimal places less than with Ethernet. Hence
> FireWire is better suited to time-critical data transmission such as
> uncompressed audio and video data. Whether the industry will ever take it on,
> however, will reveal itself in the fullness of time. 



On Tue, 23 Nov 2004 01:13:26 -0800, John Gilmore <address@hidden> wrote:
> > -max data rate ~750Mb/s, the rest is lost as signaling overhead
> 
> Actually it's better than that.  With 1500 byte packet sizes the
> Ethernet overhead is only a few percent, so well driven hardware runs
> well over 950 Mbit/sec.  IP and UDP would eat a bit more of that,
> but it's only 28 out of 1500 bytes (another 2%).
> 
> GigE always runs full duplex, which is a big advantage over USB2.  You
> get the same bandwidth simultaneously in each direction (2 Gbit/sec
> total) and they don't interfere with each other on the wire.
> 
> I'd skip using jumbo frames, since they're nonstandard and might not
> work through various switches or networks.  See:
> 
>  http://darkwing.uoregon.edu/~joe/jumbos/jumbo-frames.pdf
> 
> > -gigabit ethernet interfaces are becoming common and cheap
> 
> True.  But they don't exist on very many laptops, and it would be a
> lot more work to make it ramp down to do the 100 Mbit Ethernet PHY
> when plugged into one.  (The USRP still won't run on USB1, only USB2;
> Eric says it's too much work and he wants to clean that up later, or
> preferably never.)  I've only seen GigE on Apple laptops so far,
> though I haven't looked very hard.
> 
> > -can remotely locate hardware
> 
> Well, if you have a gigabit fiber between the hardware and the PC...
> or a 100m coax cable.  But indeed 100m is further than USB cables go.
> 
> We did look at GigE.  I think the big kicker was (1) we didn't expect
> GigE on laptops as soon as USB2 on laptops, and (2) a very handy
> PHY/controller chip was available for USB2.
> 
> But feel free to start hacking on one yourself, it would be useful!
> Make it take the same daughterboards as the USRP...
> 
>        John
>