Olimex A10S/A20-OLinuXino boards quite BBB-like

Morgaine Dinova wrote:

 

If that spec is what someone needs and the extra cost is still viable,

I can't see it, not unless you've had your head in the sand for a couple of years and don't know the RPi or BBB exist. Sure, there are philosophical reasons for wanting a more open platform than an RPi, but I always wonder how those sorts of reasons stack up against the financial cost. Sure, there are those of us who might believe it's worth it, but the masses won't.

  Unfortunately it's not sold in the UK and so they didn't get my money.  I hope that Olimex takes note of this.

I do too.  I'm still considering getting a couple of wandboards from future-electronics who seem to offer the best gbp price while still shipping from the US. Olimex are in the same boat though, while there's some limited availability here, I'm unlikely to invest a lot of time/effort building something around a board that I can't easily replace if something ever does go wrong.

While we don't seem to have had much luck with Farnell getting a better range of these boards, I wonder if someone at CPC might be listening.. Could be an opportunity for them

selsinork wrote:

 

 

While we don't seem to have had much luck with Farnell getting a better range of these boards, I wonder if someone at CPC might be listening.. Could be an opportunity for them

I Hope so   It does look like Farnell is pitching CPC  more for the "Maker" market they have produced a MakerTech catalogue.

I have driven past CPC twice this week without calling in, suffering withdrawal symptoms.

One thing that has long puzzled me is why no ARM licensee has seen fit to release a board providing two NICs, AFAIK.

Such boards are common in the x86 world, especially ATX server boards and those using the Mini-ITX form factor, but because they are generic motherboards and thus contain (as usual) everything, their prices are typically high.  So high in fact that networking enthusiasts have to resort to hacking commodity routers like the Linksys WRT54GL to obtain a dual NIC comms device at a reasonable price.

The first open hardware dual-NIC ARM board that appears at a "Pi era" price is going to have the market almost to itself, and I think it is reasonable to expect that legions of WRT users would snap it up.  Also, the A20 SoC provides 1000Mbps Ethernet capability, and it would be low-hanging fruit to harness this.

To that end, here is a little suggestion for Olimex's hypothetical OLinuXino comms and servers product line:

Single-NIC comms:

Dual-NIC comms:

(*) A20 seems not to have dual MAC capability, so replaced here by i.MX6.**

(**) It turns out that i.MX6 doesn't have dual MAC either and can't handle full gigabit anyway, so replaced i.M6 by "XXX".

The proposed pricing is just guesstimate figures intended to reflect the BOM cost savings made by omitting the HDMI and VGA support devices and sockets, since the intended application area is headless communication devices and servers.  Devices like routers and firewalls typically wouldn't need SATA either, so even more savings are possible.  As a result, some of these boards in the hypothetical comms and servers product range could in principle be cheaper than Olimex's currently listed A20 OLinuXinos.

Of course, this assumes that a dual-MAC SoC offering full gigabit performance exists at A20 prices ... a tall order!  That said, today's wishful thinking is tomorrow's reality, and these suggestions are for tomorrow.

It's a nice sunny Sunday afternoon for dreaming up products.

Morgaine.

PS. The second table has been evolving as a result of our discussions, from its original A20 to i.MX6 and now to just an XXX placeholder.  These changes may confuse any references in the thread below a bit, but if in doubt, just blame me.

Morgaine Dinova wrote:

 

One thing that has long puzzled me is why no ARM licensee has seen fit to release a board providing two NICs, AFAIK.

The AM3359 does include two NICs, and there is a board using it with two ethernet ports 21030432103043 however when you read the doc for that board, it appears that both ethernet ports are connected to the PRUs and hence you get access to EtherCat, but it's not clear if or how you would use them for normal ethernet.

It's not immediately clear to me that the iMX6 has dual mac capability either. The reference manual is huge though, so I may have missed it. Got a pointer to that bit of useful information ?

That said, as the i.MX6 has PCIe capability it's certainly possible to simply add multiple extra nics that way.

While dual, or triple, nic configurations (perhaps similar to the PCEngines ALIX boards http://www.element14.com/community/community/knode/single-board_computers/blog/2013/06/06/alix-boards-the-open-source-alternative-to-proprietary-routers ) would be nice, if you have a vlan capable switch then it's possible to get to the same place with a single nic. Although that does bring us back to the point you made in another thread - 10/100 network is no longer good enough and it appears as if the A20 is similar to the A10 in only having 10/100, although you'd be forgiven for thinking neither of them have any network capability since it doesn't seem to get a mention in any of the datasheets.

So while we're dreaming up products for Olimex, there's already an A20 MarsBoard http://www.hotmcu.com/marsboard-a20-dev-board-p-60.html

which has 1GB, SATA, Video, Flash at $58 or approx 45 euro, downside of course is the 10/100 nic.

The other interesting plus point for the A20 is that the cortex A7 is feature compatible with the A15, so you get the hardware virtualisation support. While that won't matter for a lot of people, it does open up a whole new set of posibilities.

While generally the A7 will be slower per MHz than an A9, it seems to be able to reach higher speeds which will offset this somewhat. So I wonder if in a lot of ways a quad core A7 might displace the quad core A9 designs like the i.MX6 simply due to roughly equal performance but the A7's feature compatibility with the A15 being the killer feature.

Allwinners A20 then stealing the jump on a lot of others in a mostly pin compatible drop in replacement chip certainly makes for interesting times..

selsinork wrote:

 

The AM3359 does include two NICs, and there is a board using it with two ethernet ports Re: Olimex A10S/A20-OLinuXino boards quite BBB-likeRe: Olimex A10S/A20-OLinuXino boards quite BBB-like2103043 however when you read the doc for that board, it appears that both ethernet ports are connected to the PRUs and hence you get access to EtherCat, but it's not clear if or how you would use them for normal ethernet.

Wow, awesome!  How did we miss that board among ARM discussions?  It even has quite a reasonable price!

however when you read the doc for that board, it appears that both ethernet ports are connected to the PRUs and hence you get access to EtherCat, but it's not clear if or how you would use them for normal ethernet.

EtherCAT and Ethernet are transparent to each other.  You can run both simultaneously through the same hardware, differentiated only by frame IDs.  I assume that the same driver handles both, and if the EtherCAT mode is set on an incoming frame then the driver makes it persist in the outgoing frame.

The big problem with the above board's design though is that by using the PRUs for realtime performance in EtherCAT, they've necessarily restricted their Ethernet speed to 10/100Mbps, which is a big shame when the AM3359's MACs have gigabit capability.  The board is already "interesting" just because of the extreme rarity of dual-NIC boards, but gigabit would have made it awesome.  (But unfortunately, also more expensive.)

It's not immediately clear to me that the iMX6 has dual mac capability either. The reference manual is huge though, so I may have missed it. Got a pointer to that bit of useful information ?

Good point, I wonder where I got that from.  If I find the corresponding info in the datasheet, I'll add it here later.  But if i.MX6 doesn't support dual NICs then simply replace i.MX6 by TI's equivalent SoC which most probably does.  Olimex likes TI anyway.

it appears as if the A20 is similar to the A10 in only having 10/100, although you'd be forgiven for thinking neither of them have any network capability since it doesn't seem to get a mention in any of the datasheets.

The A20's short datasheet mentions 1000Mbps in a bullet list about GMAC features on page 8:

- Support 10/100/1000Mbps data transfer rates RGMII interface to communicate with an external Gigabit PHY

It also mentions that it can handle Jumbo Ethernet Frames up to 16KB in size, which is very cool not only for fast networking but also for use in a comms device that needs to encapsulate full 1500-octet Ethernet frames in ATM for PPPoA without limiting MTUs to the dreaded limit of 1492 that causes many people problems.

With so many nice features, it's a pity that the A20 isn't aimed at communications, otherwise it would offer more than just a single MAC.  On the other hand, if it did then it would probably cost more unless it sacrificed something.

Morgaine.

Morgaine Dinova wrote:

 

Wow, awesome!  How did we miss that board among ARM discussions?  It even has quite a reasonable price!

I found it by luck many months ago. I may have mentioned it here.. somewhere.. but not sure.

EtherCAT and Ethernet are transparent to each other.  You can run both simultaneously through the same hardware, differentiated only by frame IDs.  I assume that the same driver handles both, and if the EtherCAT mode is set on an incoming frame then the driver makes it persist in the outgoing frame.

So the question is...... The BBB people say it's ok to use the 3358 (?) since the only additional feature of the 3359 is EtherCat and it's, somehow, not being used on the BBB.  So given we have EtherCat capability in the current 3359 why can't we use it ?  Is there something different about the hardware on the BBB ?  Something not wired up ?  Picked the wrong phy ?

There's something in the docs saying that the EtherCat side is handled by the PRU and not the more normal linux driver. I don't pretend to know enough about EtherCat or the implementation on the am3359, but it would appear that it's not quite as simple.

The A20's short datasheet mentions 1000Mbps in a bullet list about GMAC features on page 8:

Good catch, I missed that. So it has an internal 10/100 MAC and needs some external GMAC to do 1000Mbps ?   I wonder how many boards will bother implementing gigabit if that's the case.

selsinork wrote:

 

So given we have EtherCat capability in the current 3359 why can't we use it ?

My understanding was that we can use EtherCAT on the current AM3359 version of BBB, but nobody has so far simply because it's of such esoteric interest.  When CircuitCo shifts to using the AM3358BZCZ100 SoC instead of the AM3359, then we won't be able to use EtherCAT at all, even if we wished to.

I wanted to know what aspect of the AM3358BZCZ100 would prevent use of EtherCAT and so I asked the question here in the early days of BBB, but nobody answered with that information.

selsinork wrote:

 

It's not immediately clear to me that the iMX6 has dual mac capability either. The reference manual is huge though, so I may have missed it. Got a pointer to that bit of useful information ?

After a rummage through the i.MX6 datasheet, I agree with you, it doesn't offer dual NICs, so wherever I got that notion from, I was wrong.  As you mentioned, it does offer PCIe so perhaps a dual-NIC board could be built using external dual gigabit MACs and PHYs, but that would make the cost sky-rocket.

The datasheet did remind me though of a point you mentioned some time back, that the i.MX6's implementation of gigabit Ethernet is constrained by  a design fault which limits its maximum throughput to 400-470Mbps.  That is clearly unacceptable for most gigabit applications.  Freescale don't position the i.MX6 as a communications processor, and I think they're being fair and accurate in this.

Perhaps I should dive into TI's SoC jungle and replace i.MX6 in my Olimex wishlist table by a TI device with dual MACs and full gigabit operation.  For now, I've changed the SoC in the dual-NIC rows of the table to a hypothetical XXX.  The march of progress is sure to provide a good candidate soon.

Morgaine.

Morgaine Dinova wrote:

 

My understanding was that we can use EtherCAT on the current AM3359 version of BBB,

Fair enough, I don't know enough about it to argue.  I just wonder if there's something more to it, was John right in that other thread suggesting you need two ports ?

Also, in the ICE user manual it very specifically says "10/100 Ethernet PRU Controlled Ports". What's the significance of that statement, if any ?  The ICE board also uses a different PHY from BBB, simple cost savings or something not supported ?

For practical purposes there's no difference between all the necessary hardware being present (and connected) but no driver being available and the hardware requirements not being fully met on the BBB somehow. They are very different though, in the first case you just need to be able to write the driver or pay someone to do it..

I expect it's all academic though, fairly clear that we simply don't know enough to implement it ourselves right now - at least not without a lot of help from someone who does.