Epitaxial vs 1N4002 Diodes

• SP3002 0.85pF, 12kV, Low Capacitance Diode Array for ESD Protection  should work for ESD protection...

Clem

Many thanks, Clem.

In the meantime after a reasonable investigation on the use I had in mind it seems that it is not the better solution. I was exploring the possibility to have a passive hub but it is not so reliable and - worst of all - it seems that with the LAN works only in half-duplex. The best usage seems to be with TV signals integration but it is not my case.

Unfortunately LAN is a half duplex protocol as you found out.  If you want full duplex, SPI is the way to go. But it is meant  as a limited distance. How far do you need to go?

Clem

LAN has been full-duplex..  for some time now (*). For 10 and 100Mbps there is a pair of wires dedicated to each direction, and both can be active at the same time. For 1Gbps LAN over CAT5 cable, all four pairs are used. There can be datatransmission going on all pairs in EACH direction at the same time.

But I'm getting the impression that I'm missing some context.

(*) Remember the coax cables for 10mbps? If you got a terminator wrong, the whole network stopped working. (#)

(#) Anecdote: 1992-ish: I wanted to hook up an extra workstation in my room. So I hooked it up and walked over to the sysadmin to ask him if I'd done it correctly. But by the time I got there, his phone was red hot, so I saw that he was busy with more important things and I went back to work.... It hit me when 15 minutes later I saw him diving under my desk..... the last room he was checking for missing terminators and the like....

I am referring to the protocol at software level. Cannot send/receive at same time in software interface. But there is some acks under the cover that are overlapping.

Clem

That's a limitation of the application. Not the network stack or you would never be able to surf in multiple browsers and have a file transfer going and watch TV / my youtube videos at the same time  

I'm baffled now - are we seem to be talking about Ethernet but which protocol:

It's easy to send and receive UDPs simultaneously, TCP links usually look full duplex even to quite high level software - I'm obviously missing something but not sure what

MK

clem57 wrote

Unfortunately LAN is a half duplex protocol as you found out.  If you want full duplex, SPI is the way to go. But it is meant  as a limited distance. How far do you need to go?

Clem

I have replied from my mobile to this comment but now I see that it seems nothing happened, so I repeat. What I plan is to make a inter-network between 2-3 raspi in the same place (I mean almost the same box), so the distance is few centimetres less or about 20 cm every connection. Then for any networked external connection with the resulting device set I have to use a common switch or any other reliable network.

What do you intend as SPI ?

Why bother with the diodes then if this is so close together.

If you want them for some special reason you really need to state the environment  otherwise there is little benefit in the answer as every one is taking a guess at whet you are working towards.

Also yeah Ethernet is Bi-Directional Duplex

John

Why bother with the diodes then if this is so close together.

If you want them for some special reason you really need to state the environment  otherwise there is little benefit in the answer as every one is taking a guess at whet you are working towards.

Also yeah Ethernet is Bi-Directional Duplex

John

John, the - possibile - choice of the passive small hub with diodes only was to ensure data exchange between the processors using the Ethernet protocol (that simplifies a lot of things in kind of communication I am approaching) but I don't understand what do you mean with your suggestion, what kind of alternative is ?

I, and many others I guess, have no clue whatsoever what you think you can achieve with diodes on ethernet. To connect several pi's together you need a hub or switch. A hub will transmit on all outgoing links as soon as it recieves something on one of the incoming RX lines. A switch will quickly analyse the incoming transmission and transmit only on the appropriate port (with a time-delay of a few microseconds). In the past (say more than 15 years ago) you might chose a "hub" over a switch because of costs. As the price difference has dropped, I don't think hubs are in serious active production anymore.

So, besides your "four pi's" you'll need a small switch. 5port 5 Ports 10 100Mbps Fast Ethernet Network Switch Hub USB Cable Kable SY | eBay

Or you could connect the SPI ports together. The problem is that SPI is a master slave protocol, and the "pi as a slave" part has not been implemented yet. You'll need to do that yourself. Big job. Same goes for other options. Unless you have very specific needs, you're better off just putting a network switch near your 'pi's...

Roger,

the point is that I know very well both hub and switches roles and how do these devices works, collision problems and so on. Instead I have not a sufficiently wide visibility about the concept of "passive hub", that is, a unpowered hub to connect three different RaspPI. The thing is possible, and a circuit exists but the logic on it is based is not convincing me very well. (here is on link: http://www.diyelectronicsprojects.com/2012/07/building-passive-ethernet-hub.html and all the hundreds of others I have found are identical to this scheme, no matter of the site).

Instead frankly I had not considered the SPI connection that will be a good idea, excluding the problem you mention of the unimplemented slave mode.

I see that there is a master-to-master SPI connection solution but it seems having only one wire for data and this slowdown again the speed. The idea for a direct connection via ethernet has a lot of advantages in the SciFi project (tag: meditech_project) I am developing and the solution that has sense in a modular environment with the ability to connect the Raspberry-pi modules in the same place or remotely is just the ethernet way. As there will be two kind of networks, one barely local to exchange the data flow and one to the Internet via the WiFi dongle, I think that the final solution I will adopt is an internal hub.

I was exploring this part of the project also because this choice influences both the contained design of the portable device and the powering system.

Just buy a small 5-8 port switch and stay at a nice 100Mbps and find your are much aster because most switches will switch at near line speed and you wont have any  collisions as such as you have more control of collision domains

More to the point who would want  to use the RPI for low latency switching when the USB to Ethernet  is so notoriously slow and fickle anyway !?!

Now you have explained what you want to do and referenced the basic circuit it is possible to answer your questions.

1) The whole idea is pretty dreadful since it relies on undocumented and unintended behavior of of the 10baseT or 100baseT hardware - however it obviously could work in some setups.

2) The diodes will need to be fast - 1N4002 would not work, 1N4148s probably would.

3) You can buy a 5 port switch from Amazon for £5.86 - you won't be able to buy one switch chip and the connectors for this price so building your own switch would be mad.

Your best bet (by far) is to buy a cheap switch and build the works of it into your system.

But pay attention to John's last remark - if you care about Ethernet performance the RPI is the wrong platform.

IMO you can remove the condition from that last sentence

What is your project - we might be able to suggest some alternatives to allow three processors to have a low latency network.

MK

Agree with you on all points there.

Need to know what sort of project he is taking on to give some idea of required speed and acceptable latency!

Sorry for the delay in my answers, I have read all your last comments on the mobile that is not very good to write decent responses.

John, the problem is not the cost as I already have one to make tests. The speed of 10/100 is more than sufficient. Your answer and the other guys before already gives me the correct vision, despite the project itself, that the choice now is between ISP (that is the less valuable I think) and the LAN connection that it is very probably I will decide for.

Michael, as you wrote

1) The whole idea is pretty dreadful since it relies on undocumented and unintended behavior of of the 10baseT or 100baseT hardware - however it obviously could work in some setups.

2) The diodes will need to be fast - 1N4002 would not work, 1N4148s probably would.

3) You can buy a 5 port switch from Amazon for £5.86 - you won't be able to buy one switch chip and the connectors for this price so building your own switch would be mad.

1) That's ok, I have explored again the of the passive hub (mostly for the power consumption but it is unreliable for too much aspects, so it is excluded apriori

2) This is anyway an information that I need at least for curiosity; following simply the logic, if there are specific high speed diodes there would be a reason, and this is another point that address me to exclude the solution.

3) That's ok for the mini switch and the problem is not the price itself

The project is - in general - metitech_project (this is the tag, you can find all the already wrote posts in the SiFi your Raspberry-Pi challenge blog). In this moment I am focusing all the obscure points, ad this is one of them, to define the complete architecture of the system.

For several reasons I have to use more than one RPI and I should have them exchanging data together in a private internal connection. Then at least two units will be in the same "box" or container. The system in general will be modular so some complex probes - better group of probes - will be external pluggable components driven by another RPI. The same mobile lightweight box will be also available in  a non-portable version with the devices connected together via a network. The "main controller" RPI will include a wifi for the main interface (Android based) so with these premises I think that the LAN is revealing to be the right solution.

The doubt that I already have is about the suggestion os clem57 about the possibility to use for the internal connections the SPI protocol. As I wrote in a precedent post, if by one side the SPI protocol on RPI has the limitation that doesn't support the slave mode by the other there is the master to master connection that - as for what I have read - seems to use only one channel instead of the traditional two bidirectional wire lines MISO and MOSI.

Anyway a t the moment I am strongly prepending for the LAN also for the internal connections.

Hello Enrico,

Now I know about your project I would definitely use LAN for communications. The advantages of SPI would not be available to you in this context. The implementation of Ethernet on the Pi is far from ideal but will probably be OK (I'm assuming that the earlier reliability issues have been sorted out  - some RPI expert may have  a comment here.)

One big advantage of using Ethernet is that the integration with Linux is far better than for any other reasonable bandwidth communications available to you on a PI.

MK