From: J G Miller on
On Thu, 15 Jul 2010 14:30:35 -0500, Grinder wrote:

> It is some sort of cosmic justice that people making snotty replies
> often are making errors of their own.

Sorry if I sounded snotty -- the last point I was making is that one
should really should show some effort at having done some searching
and declaring that something is not available when a quick search
reveals that there is something similar or provides the same
functionality is a little perturbing.

> To the OP: I have yet to see that device, and have only seen complaints
> reported concerning its USB 2.0 analog.

The cable for which I provided a link was the USB 2.0 version.

I was pointint out that such a cable existed, not how well it may
or may not work.

So what are the complaints about this cable which you have heard?

I am most intrigued.
From: J G Miller on
On Thu, 15 Jul 2010 13:04:36 -0700, John explained:
>
> Because I want to benefit from the 5 Gbps transfer speed of USB 3.0

At what speed are you reading and writing the data on each machine?

> Your idea would require me to purchase 10 pieces of 1 Gbps NICs

Why would you not just buy one 10 Gbps NIC for each machine?

<http://www.myri.COM/Myri-10G/10gbe_solutions.html>

> to create one virtual 5 Gbps ethernet interface.

I may very well be incorrect, but I doubt very much that you are
going to get 5 Gbps *ethernet* on a USB 3.0 5 Gbps cable.

> Are you joking?

No, trying to ascertain what you want and why you want it.

> And that cable you linked to is USB 2.0 which is limited
> to 480 Mbps

Yes, because there is very little USB 3.0 equipment available
*yet*. But it will arrive in due course.

> Maybe not.

Maybe not, but you gave no indication of having done any
searching.
From: Paul on
John wrote:
> I want to connect two computers with USB 3.0 and run IPv4 over USB.
> Seems to be easier said than done. How can I make a PC into a USB
> device? I can't find PCI cards or similar that will make the PC act as
> a USB device rather than a USB host. Do I have to wait until someone
> is kind enough to make a cable with a device-host-device in the
> middle? Surely there will soon be such a cable on the market, but I
> doubt it will run at 5 Gbps... which is the entire point since I don't
> want to invest in 10 Gbps Ethernet of infiband.
>
> Any ideas are much appreciated :)
>
> /Z

Everyone wants to beat the commercial pricing of high speed Ethernet
adapters. I haven't seen a "USB3 ping-pong" chip yet, to do that.

If any company was going to do one, it would be these guys.

http://www.prolific.com.tw/eng/Products-2.asp?ID=10

They claim to have already done one other chip, with a
USB3 PHY on it, so that means they have all the ingredients to
make a USB3 ping-pong chip if they want to (or see a market
for such). My guess is, this will be further down their
"to-do" list, as there isn't enough product volume to
justify rushing one out.

USB is a host to peripheral technology. The "ping pong" chip,
allows host to host communications, by faking it (making each
computer think the connection is host to peripheral). The
innards of the chip are pretty simple, consisting of a couple
unidirectional FIFOs ("mailboxes"). At USB3 rates, that structure
may need some adjustments, to lead to high performance. Otherwise,
it could have a relatively high CPU overhead requirement
(data copying and the like). In fact, without care, it might
even be CPU limited and not make it all the way to 400MB/sec.

Paul
From: John on
On Jul 15, 10:47 pm, J G Miller <mil...(a)yoyo.ORG> wrote:
> On Thu, 15 Jul 2010 13:04:36 -0700, John explained:
> > Because I want to benefit from the 5 Gbps transfer speed of USB 3.0
>
> At what speed are you reading and writing the data on each machine?

At the speed of DD3 RAM, i.e. more than 5 Gbps.

> > Your idea would require me to purchase 10 pieces of 1 Gbps NICs
>
> Why would you not just buy one 10 Gbps NIC for each machine?

Because they are expensive.

> I may very well be incorrect, but I doubt very much that you are
> going to get 5 Gbps *ethernet* on a USB 3.0 5 Gbps cable.

So do I, but I would like to have facts before I dismiss the idea.

> > Are you joking?
>
> No, trying to ascertain what you want and why you want it.

Ok, I thought it was obvious that I wanted high performance since I
wrote in the original post that the alternative was 10 Gbps Ethernet
or infiband.

> > And that cable you linked to is USB 2.0 which is limited
> > to 480 Mbps
>
> Yes, because there is very little USB 3.0 equipment available
> *yet*.  But it will arrive in due course.

What I just don't get, and would like to know, is why there aren't USB
device chips that plug in to the PCI port of a PC.

> > Maybe not.
>
> Maybe not, but you gave no indication of having done any
> searching.

That is quite true.

All the best,
Z
From: John on
On Jul 15, 11:15 pm, Paul <nos...(a)needed.com> wrote:
> Everyone wants to beat the commercial pricing of high speed Ethernet
> adapters. I haven't seen a "USB3 ping-pong" chip yet, to do that.

5 Gbps (or slightly less) would be perfect... there really is a window
in time for that now. Few people really needs 10 Gbps, quite a few
want more than 1 Gbps.

> If any company was going to do one, it would be these guys.
>
> http://www.prolific.com.tw/eng/Products-2.asp?ID=10
>
> They claim to have already done one other chip, with a
> USB3 PHY on it, so that means they have all the ingredients to
> make a USB3 ping-pong chip if they want to (or see a market
> for such). My guess is, this will be further down their
> "to-do" list, as there isn't enough product volume to
> justify rushing one out.

But the kind of peripheral chip that sits in USB 3.0 portable disk
drives on the market now, why can't one take that kind of chip and put
on a PCI-express card that plugs into the PC? And then let the PC/PCI-
card act as a peripheral?

> USB is a host to peripheral technology. The "ping pong" chip,
> allows host to host communications, by faking it (making each
> computer think the connection is host to peripheral). The
> innards of the chip are pretty simple, consisting of a couple
> unidirectional FIFOs ("mailboxes"). At USB3 rates, that structure
> may need some adjustments, to lead to high performance. Otherwise,
> it could have a relatively high CPU overhead requirement
> (data copying and the like). In fact, without care, it might
> even be CPU limited and not make it all the way to 400MB/sec.

So it's even more unlikely that we will se a USB "switch" that makes
it possible to create a small high-speed LAN? Then all peripherals
must route their packets through the PC acting as host?

All the best,
Z