I recommend you some literature on this topic. Embedded real-time systems are not data centers.
Hermann Kopetz, Real-Time Systems: Design Principles for Distributed Embedded Applications, Springer, 2011, ISBN: 978-1-4419-8236-0 Roman Obermaisser, Event-Triggered and Time-Triggered Control Paradigms, Springer, 2005, ISBN: 0-387-23043-2 There is an Raspberri Pi model with PCB edge connector: https://www.raspberrypi.org/raspberry-pi-compute-module-new-product/ But I don't get it. What makes this type of connector better than a pin header? If one wants to connect sensors to this device it would be necessary to create a break-out board anyway. Rafael writes: > Perhaps it's just semantics but we are not making progress here ;-) > > On 06/06/2015 05:46 AM, Alexander Rössler wrote: >> >> Rafael writes: >> >>> On 06/05/2015 01:18 AM, Alexander Rössler wrote: >>>> >>>> Rafael writes: >>>> >>>>> On 06/04/2015 07:13 AM, Ron Bean wrote: >>>>>>> If you need one computer to see the GUI and one for realtime >>>>>>> effects, why not just start out with a real computer and load Linux and >>>>>>> LinuxCNC on it? >>>>> >>> .... snip >>> >>>>> In my HW support experience I came across PDP-11 systems running in >>>>> steel mills, nuclear and hydro power plants, factories, etc. with little >>>>> or no graphics. Most used VT100, some used more advanced color >>>>> terminals. Systems with 32kW(ord) or 64kW RAM controlled huge machinery >>>>> with RTOS on much slower CPU than we have today. >>>> The future are distributed systems. Distributed setups are industrial >>>> standard and are used everywhere from automotive to automation >>>> industry. CAN and Ethernet are used these days to distribute >>> >>> neither one is suitable for strict real time. >> CAN as event triggered bus is not. You may understand TCP/IP as > Huh??? > > "BUS" is not event triggered? Every bus I know has an interrupt line(s). > Generic BUS is just a data path, not an interface, you are mixing the two. > >> Ethernet. However, Ethernet can be used as time-triggered bus too. There >> are many standards such as EtherCAT and Powerlink which are widely used >> in automation industry. > > But that is not a data bus I'm talking about! Using your definition, > RS232, parallel port (see, we call it a port), phone line, and > traditional CaTV network are also a bus. > > In my understanding BUS is physical component of a computer to connect > numerous peripherals to the CPU and among themselves. DMA for example is > used for data transfer between the peripheral and storage (RAM, disk > drive, SSD) without CPU intervention. No such thing in ethernet. > > All these concepts were resolved on mainframes decades ago. What changed > is the size of components and their speed. > >>>> functionality across different ECUs. The BBB is fine when it comes to >>>> CAN but an even stronger platform from TI is coming up: the BeagleBoard >>>> X15 with Gigabit Ethernet support >>> >>> Don't mix computer BUS and cabling. Two different things. Some cables do >>> act as traditional extend bus but none at the length of an airplane or >>> HMMVE. >>> >>> What good is Gigabit Ethernet when you need to connect a keypad, a >>> switch, accelerometer, or optical sensor to BBB? Ethernet is not a bus, >>> it's one of communications peripherals. >> You are wrong, Ethernet is a bus. When you take a look at the history > > This is becoming silly. Ethernet IS NOT A computer BUS just like USB, > RS232, RS485, 60mA current loop, are not traditional data buses. Perhaps > the terminology got confused when USB was introduced and you know how > much of a bus that is or how real time that can be. > > I work in large data centers and never hear anybody calling ethernet a > bus. In some instances we use Cat-6 cables, in other we use optical for > 10Gb or 40Gb connections between the routers, switches and servers. I > would never call that cheap. A single ethernet interface costs more than > a large box of BBBs. > >> you will see that it started out with a very different physical >> interface as nowadays. The huge advantage of Ethernet is that network > > I know very well how it started and what kind of connectors were used to > connect ethernet interfaces on DEC or other computers over the years. At > some point coax cable was used to string PCs together into a network. I > probably still have RG-58 cables around, good for amateur radio. > > If you forgot 50 ohm terminator, or used a cheap BNC connector, that > connection went "wireless" and transmitted all over the spectrum except > between the computers. I paid my share to that hell but we ever called > it an ethernet BUS! > >> hardware is cheap (not all is RT compatible though) because it acts only >> on the data-link layer (Ethernet frames). What I am talking about are >> Ethernet hubs. > > I hear "hardware is cheap" all the time. It is in some instances but > when I ask an engineer that demands additional disk space because he > thinks hardware is cheap: "if it's cheap, why don't you go out and buy a > disk drive?" they walk away and start cleaning their home directories > full of "junk" on servers because management tells them to do so. > > Cost is always relative to age, volumes, and features. > >> The idea of time-triggered buses is to resend that every > > resend what? Resending packets to fix broken blocks of data is very > costly. Idea is not to resend anything. Send it once and be done with. > You cannot afford to lose an interrupt when a mill is at the stop switch. > >> cycle. Therefore, a higher network bandwidth means that one can use a >> smaller cycle time. The bandwidth is not wasted as some people stated. >> > > Ethernet packet is not guaranteed to make it to the other side, speed is > not an issue. If you connect only two devices you may get away with it, > add a switch and you have completely different scenario. And it gets costly! > >> Why not attaching the sensors you mentioned directly to the BBB? Just > > Where I've said that? You still need to connect to a board of some kind, > call it shield or whatever. But ... > >> create (or use one of the many) capes with a decent connector and you >> are fine. > > you seem to completely miss the point. Completely. My original post was > related to physical awkwardness of "little computers" with non standard > BUS/connectors/protocol. > > It's possible to build interfaces (shields, capes, or whatever other > silly names they come up with) for any computer if you want to go one > architecture of a kind but it's costly. > > You don't need to go further than > http://linuxgizmos.com/category/boards/ to see the SBC chaos. Many > manufacturers don't want to be compatible with a product from another > company. It's similar situation as "Personal Computers" in the 80's. > First "personal computers" from DEC were not compatible with HP, IBM, > ATT, and bunch of others. > > Try to find interfaces that you can exchange between "hacker SBCs": > http://linuxgizmos.com/vote-now-for-your-favorite-hacker-sbcs-maybe-win-one/ > > This emerging standard is the closest to what I'm talking about: > http://linuxgizmos.com/new-arm-and-x86-com-standard-gets-a-boost/ but > the cost is still relatively high. If those who design capes, shields > etc. made them for a standard like SMARC you would see more affordable > prices for hackers. > > Just in case, computer data bus is not dead: > http://linuxgizmos.com/arm-based-device-developers-get-smarc-coms/ > >> If you want to go the industrial standard way you can buy sensors with >> bus interface (I am not talking about I2C, SPI, ...). Onewire is common >> for simple sensors. Another example in the automotive industry it is >> pretty common to have ECUs that do only simple tasks like reading out >> sensors and providing the data on a CAN bus. With microprocessors >> getting cheaper and cheaper the industry will further move into >> distributed systems. > > Again, the discussion was not about what kind of sensors you can buy but > interfaces and their physical characteristics in relation to specific > SBCs: Arduino, RaspberryPi, etc. > > Go to Maker Faire and you can see many different "little boards" for all > kinds of things but they are mostly not compatible with each other. But > they are cheap. > >>>> On the other edge of the spectrum we have another low cost solution that >>>> is currently funded on kickstarter C.H.I.P. a 9$ dollar Linux computer >>>> with Bluetooth and WLAN => a cheap solution to connect sensors. >>> >>> This is one of a kind toys that don't make a standard! Nor would anybody >>> serious use it for a CNC machine. >>> >>>> I even heard about things like fly-by-wireless. Which boils down to >>>> removing the wired buses inside a plane. So face the facts: Big >>>> monolithic computer setups will soon be banned to server farms. >>> >>> Most airplanes and modern military vehicles use computers based on >>> decades of developments on VME bus and it's derivatives because they >>> need a lot of connections. That likely includes CompactPCI, it's >>> emerging CompactPCI Serial, and VPX. >>> >>> As tiny lasers are getting cheaper, cost of building optical bus and >>> compatible peripherals will become more common in the near future so >>> we'll see even more data buses. >> Optical cables have different problems than metal cables. They have more >> problems when it comes to mechanical stress. I am not sure they will >> succeed copper wires that quickly. > > I'm working with optical cables on a daily basis for some time now. > Again, you don't understand what an optical bus is, it is not a cable, > which you seem to mix every time subject of data bus is coming up. > Ethernet over optical lines is not a bus. > >> When you take a look inside an airplane you will see that the wiring is >> consuming a lot of space inside the hull. The idea of replacing some >> buses with wireless interfaces drastically reduces development costs. So >> maybe in 30-50 years we will have wireless operating planes. > > On my tour at Boeing I learned that wiring is not one of the major problems. > > Wireless is not magic to solve all of your wiring problems. Radio will > never replace wiring in "mission critical systems". Airplane is one of > them. Man, that would be simple way to get airplanes off the sky! > >>> Every computer in existence has a bus, available or not, for connections >>> to additional peripherals. There is a bus on BBB, RaspberryPi, Radxa, >>> and other little SBCs to add peripherals. My comment was about the >>> problem with every little SBC having different connectors and their >>> positions on the board while all are using "sandwich mechanical >>> architecture" that cannot be expanded easily. >> What you are pointing out is that these devices do not come with >> standard connectors. There are some capes (additional board that can be >> put on the pin headers) that provide different connectors for different >> applications. The BeagleBone Green will come with connectors for the >> Groove sensors if you want something out of the box. Furthermore, you >> have USB and Ethernet connectors available. > > Can you use Arduino ethernet card on BBB? Can you use wireless card from > one platform on another? No and that's my problem. > >> However, I agree that connectors are a big problem in general when it >> comes to computers. Only few capes address this problem an come with pin >> headers to connect sensors/motors. However, that is one problem we tried >> to address with the SandyBox and the different controller boxes >> (Lin-Ctrl stepper driver, Print-Ctrl for the 3D printer) . They come >> with standard Molex connectors to connect sensors, switches an motors. > > It's here that makes me think that you are involved in SandyBox design > or production. Good for you. It's open discussions like this that can > make a big difference even though we strayed into computer history, > production, etc. > > I never commented on SandyBox specifically. My comment (again) was > related to lack of standards for hobby or DIY makers. That's all. > >> We are planning a future version of the SandyBox to address this >> problem. So if you have ideas please share them. > > I'm glad to see somebody design a box that serves it's purpose > especially if it's able to run Linux. If I understand the description > correctly, it's specifically targeting (CNC) systems that use parallel > port. There are other options for different CNC machines too many to > list them here. -- Alexander ------------------------------------------------------------------------------ _______________________________________________ Emc-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-users
