Some tips from a former embedded developer. The PHYs will also need a clock. But if they use the same frequency (e.g. 125 MHz) you can reuse the same oscillator and use clock buffers or daisy chain the clock signal. Saves you another chip. Also there are oscillator chips with multiple output clocks and you can program every output clock before hand. Another important thing are LEDs, they are useful for debugging but also have their use during production and can show if some software component is running or not. Just make holes in the case and add walls to block light bleeding. People will find many uses for the LEDs.
Former? What made you quit/leave? Thanks for the suggestions regarding the clocks, current candidate is 6V49205BNLGI8 (and 9DML0441AKILF for the buffer). You have just made me think about the LEDs though, I personally hate them anywhere else but the ethernet ports (so that I can see activity). The initial plan was to have one LED that blinks while booting up and then remains lit during normal operation. However, putting some extra on PCB for debugging purposes makes a lot of sense. Thanks!
I definitely agree with adding leds. I recently finished an embedded design where I was required to show all the power, status, and error information with one RGB LED. It took a lot of development time to figure out how to squeeze all that information through one LED when it would have been dead simple to have separate LEDs for each function. If I was doing that design again, I would use separate power, status/activity, and error LEDs. Unless PCB space is *the* limiting factor, I would at least add pads for indicator LEDs. If you don't use them there's no wasted cost, but if you need them later you can do light pipes to the case which was easier then I thought it would be I like dual color red/green leds for status and error since you can do red/green/both (amber)/black (off) flashing for intuitive severity feedback. People can guess that fast flashing red/black is clearly worse than slow green/amber. Just my two cents having flashbacks (pardon the pun) to working on UI with LEDs
@@tomazzaman in case of LEDs you may want some buttons, too. you can turn the aditional LEDs of, and leave only power LED in a dimmed state, on button press, you activate the rest of the leds to display certain status info, like sfp status an online indicator, an vpn indicator, or anything else a user can imagine. maybe take a look at the Speedport Smart 4 from Deutsche Telekom and use an OLED Display. There are multiple use cases for an display. heartbeat, actual amount of traffic, load, number of active (WiFi) clients, IP Adresses, CPU Load with an display you are not limited by the amount of leds and also can utilize it as an led replacement.
IMHO the deciding factor for buying this router would be mainline kernel support. The number of SBC style boards I've purchased over the years which lasted < 2 years because I wasn't able to update the software on them anymore because the vendor used the chip manufacturers custom kernels to build forked variants of whatever os/distro.
Not a huge fan of the external power brick, but it makes a lot of sense, heat, space, etc. Also if you were to go internal people would want redundancy, it’s a losing battle. That said, as another commenter said proper implementation of USB C PD would be nice!
Devices this small always have external power bricks though, and they are extremely common for routers and other gateway type devices. I see nothing wrong with it. Though as @lachlanstone282 said, it would be very cool if next to the DC input port you had an RJ45 port with the communication lines disconnected and only use the lines that poe+ power comes in on. It would be really nice to use a port on the main network switch that is either poe+ or poe++ and power the router like that for those of us who had the capability. But in the end, that is just a cool idea that would likely be used by few people and add to the cost.
Yes, those PoE-powered switches are great! I agree with your points, though. I also agree with what you are saying about power bricks; the exception is for rack-mounted devices. @@LordSaliss
The power brick makes it easier to launch a product else you're going to go through a cert route with added costs for 110/230VAC EMI and stuff and now you just say use a usb-c power brick.
I think what you're doing is really great. I only just found this channel and have started going through your router designing videos. Really enjoying it so far. BUT in my opinion, not having dual SFP+ is a total deal breaker; For me and probably for a lot of other people too. Especially those using fibre networks. So considering the CPU's limitations, it makes much more sense to have dual SFP+ instead of 1xSFP+ and 1xRJ45 for the 10Gb ports. That way SFP+ fibre can be used for both WAN and LAN. Then if the user does need RJ45 they can just use an RJ45 Transeiver. I think not going dual SFP+ will reduce the market that your router will appeal to. I also think that with dual SFP+, your router will have the potential to be marketed as a professional router to the SME (Small and medium enterprise/business) sector, not just homelabbers. And if priced correctly could be extremely competitive in the small-medium business market. Even against Cisco and Ubiquity's offerings. You could have a router that is damn near perfect in my eyes.
@@tomazzaman consider the ryzen SoC, you get 2x10Gb/s interfaces for embedded and a vega GPU but this is with their gen 1 ryzen embedded. Might have newer variants as mine is a bit old. This lets you easily have SFP+ as with SFP+ you can reduce some of the costs associated with PHY ethernet, however this puts the burden on customer instead for ethernet. However SFP+ direct is cheaper than 10Gb/s ethernet in short distances with ensured cable quality from my experience. After a long time running ethernet sometimes they tend to drop in speed which is annoying. So with 2x SFP+ one may put GPON opticla on one, and on the other SFP+ direct to a SFP+ switch. In my previous setup i had a 10Gb/s capable switch with 36 TILE cores with 2x SFP+, one of it to a SFP+ switch which i combine to ports to another switch before breaking it out to 24 ports. On the SFP+ switch itself i also connect file servers. I can help you find better SoC options for your project than what you are currently using. However check out ryzen embedded SoCs first as well as other ARM options. Have you also tried snapdragon given they have good OS support? intel atom is very slow in comparison to rockchip and better options.
@@tomazzamanIgnore 1/2.5gb ports and go 10gbe all the way. You give users the option of using different connections if you split up the ports evenly (3xSFP+/3×rj45).
In my experience, sfp+ is more flexible and reliable than the rj45 variant. So better use 2x sfp+ instead of 1 sfp+ and 1 10gbit rj45. When you really want rj45, you can always add a sfp+ rj45 module.
Agreed. This is incredibly important to make sure all use cases are covered. As far as I'm concerened, not having dual SFP+ is a total deal breaker for me. Personally I think having dual SFP+ should just be the default and only variant. If users need RJ45, they can just use adaptor transeivers.
My thoughts to. SFP+ modules for fibre or copper are fairly cheap now (sub $20) and come in both 1 and 10G options. Doing this lowers the cost of the unit while allowing the user to pick what they need. Would definitely be a requirement for me.
Agree with this too. 10G over copper tends to be more of a backward-compatibility use case, and it can still be supported as such with a 10G-BASE-T SFP+ module.
8:19 I think of it this way, I say out, not oot.... So R-out-er, not R-oot-er ;) Loving these deep dive videos as you work all this out. Looking forward to the rest of your journey.
For the uart , please consider using a standard RJ45 rolover type console port in addition to the type-c console port. ive had terrible luck with built in uart -> usb whereas with rj45 i have a pile of adapters that i know work every time.
I'm excited to see this project develop. 3 things I'd like to see, rackmountable with pigtails, 2x sfp+ rather than 1xsfp+ and 1x10gb, power input on the back. Good luck!
@@tomazzaman I like the idea of 1xRJ-45. The SFP+ to RJ45 transceivers generate a hell of a lot of heat. Also here in the US we get 2G fiber but the ONTs still connect via RJ45. You have to cover a broad audience. Not everyone will connect direct to SFP+ and not everyone will connect at 10G. Most of us are just getting into 2.5G. Also Rackmount option!
@@harihoudini Ubiquiti dont want to support 2.5G SFP modules. Their fans are fanatic and i wonder why people still want to pay their prices for bad options. Other brands already support it. The support is more down to software though. Since datacenters dont use ubiquiti they have no one to scream them into action meanwhile if mikrotik doesnt support a feature many are going to scream at them as businesses and datacenters do use them. Some ARM SBCs with 16GB of ram for $150 faster than intel atoms are coming with 2x2.5Gb/s ethernet.
@@tomazzaman1 x 10Gb RJ45 and 1 x SFP+ are perfect. Wouldn’t even need the 4 x 1Gb RJ45. Just need a router with 1 x 10 Gb RJ45 to connect to the Fiber modem and 1 x SFP+ to connect to a 10 Gb Switch. And a rack mount option. That is all I am looking for. Didn’t find anything reasonably priced so far and have high hopes in you project. Anything extra only makes it more costly and I hope it stays below 600-700 €/$.
@@tomazzaman Count me in the "2xsfp+ = interesting, 1xsfp+ and 1 RJ45 = no thanks" camp.. as may other said from sfp+ to rj45 is a fairly cheap device (addon option perhaps, when selling?)...
I’m very interested in this project. I speak US English and while it’s commonly pronounced both ways I prefer router, not “rooter”. I prefer an external power brick, as long as you use something standard like usb-c PD or a 12V 5/2.5mm barrel jack. This benefits replacement of a bad power supply, minimizes package size & heat dissipation, and greatly simplifies regulatory compliance (CE or UL).
You should definitely provide some way to connect 12v power through a DC barrel jack or possibly 48v if you want to support POE. Pretty much all networking gear runs on those voltages / connectors, and it's much easier to supply power at higher voltages. If you have USB-C power input you would need the switching electronics anyway because USB-C PD runs at much higher voltages (usually 20v, but it's negotiated by the charger and device) in order to deliver 10s of watts of power over relatively thin wires. Buck converters are relatively small and inexpensive these days, so you can internally downconvert it to 5 or 3.3v as necessary.
Im curious to see the OS implementation plans, it seems like opnsense has some unofficial support for ARM64 but it will likely need the latest branch of BSD to be ported to your platform which sounds costly. Hardware plans are awesome and im excited to see where it goes!
Thanks! I'm currently in talks with another vendor of telco gear that can lend me some ARM-based PCBs for me to try and test how difficult it is to port opnsense to it. Will report here, of course :)
Given the security implications of being a router in a network, did you consider: - adding UEFI secure boot with a PKI where you can give out certificates for "official" distro maintainers for your router that you trust, for end-users that still want something powerful but don't want to be a "maintainer of a router" so to say? - are you planning on using any hardening on the OS side like SELinux? - for example i'm running a raspberrypi with fedora-iot which uses rpm-ostree, systemd with podman rootless containers, etc. I'm currently looking into SELinux to harden it even more. It's nice because with 'systemd-analyze security' you can also easily analyse a lot of security related knobs for your containers/services. Good luck and thanks for all your awesome contributions to us (videos, knowledge, hardware, etc.)
Have you considered going with dual sfp+ ? SInce RJ45 transceivers would allow for dual function. It would also be nice if one of the 1GBE ports was a SFP port, since that would allow one to not have to use a ISP CPE. USB-C PD would also be very nice.
Like @omegatotal said, RJ-45 SFP+ modules get VERY hot (I have one in my switch, can attest to that), so I'd rather have the flexibility. Putting two side by side would be a risky move. Don't get me wrong though, it would be cheaper to manufacture as we'd get of one PHY chip which costs around $30 on its own, but I'd rather not force potential customers to buy more stuff to get it working.
I and almost all installs we do use fiber either OS2 or OM4 we would rather have 2x sfp+ ports seen as we run fiber between all other gear and fiber comes in in fiber format so the rj45 would be un used and that way useless or were are force ro run a sfp+ rj45 on the switch side
A huge thing for the potential cost is the longevity of the device. With an arm SoC you end up relying on someone maintaining a specialised kernel. An embedded x86 processor would give buyers the knowledge that the device can be repurposed in future and isn’t locked down (could run pfsense etc). I’d also suggest 2x sfp+ and 12v barrel jack. The 4 gigabit ports seem like a waste IMO, anyone forking out for this will likely have a core switch behind it, it may be better to offer a PCIe slot so buyers can choose their own expansion, if they want 4 gigabit it 2.5g or 10g it’s up to them.
Thanks for the feedback, apart from the ARM SoC I'm addressing the rest in my pricing video. We've indeed moved from 1x4Gb to 3x2.5Gb (because we can do that with the same CPU). Also 2x SFP+! :) x86 unfrotunately brings too many problems for a small startup to solve. And ARM CPUs can just as easily be repurposed. Yes, the support is smaller, but they are gaining in popularity, so it'll get better with time.
I really like the progress you are doing. I think that it would be nice to have redundant power inputs if it would not ad that much to the cost. It might be cool to be able to plug a PD capable powerbank to one of the power inputs as a UPS in that case.
For total of 35W (11:43) of power a 5V over single USB Type C input is not enought. Even high-current cables max out at 5A (5A*5V=25W, good luck with chargers which even support that, most likely 5A could only be used at 20V), standart current cables max out at 3A. I would recomment using 12V or 15V USB PD (power delivery) input for 36W (12V 3A) - 45-75W (15V at 3A or 5A if you are lucky)
Can you make a router that is upgradable. Upgrade radio Upgrade RAM Upgrade the main board. So this will make it better to not sell the entire router everytime one needs to upgrade, the router ‘s robust amazing aluminium can be kept as it is.
2.5GbE is only possible on one port because of the CPU limitations. I mean, I'd love to choose better components, but then the device will become much more expensive and inaccessible to some. My dad had an aluminium foundry, so I helped him a lot growing up. Because of that I signed for mechanical engineering middle school, but I partied too much so my grades weren't that good. Went to Biotechnical faculty to become a wood technology engineer, but didn't finish as I got kids and fell in love with software development. In 2012 I started a startup called codeable.io (with a partner), left last year. Now I do hardware. Or, I'm trying to :)
@@tomazzaman it make sense, given the CPU limitations. On the other hands, we are getting cheaper 5x2.5GbE + 2 10GbE (unmanaged) switches that are super affordable. I would gladly pay more for something with more capabilities. Maybe even PCI-E expansion slot as on Zima board, so one can choose the expansion. But I guess that's even more complex and expensive on the CPU side. Can't wait for the next video and some rough estimation on the pricing. Cool that you are able to pursue what interest you. Wish you all the luck with the project(s). My dad has a car/auto-service, so I started with modifying cars, then racing, now sim racing online. I studied computer science on a mathematical faculty (PMF) so I naturally went into software. But I'm so interested in hardware, IoT, smart homes and all of DIY stuff. That's why I find this channel super cool. Kudos!
@@tomazzaman before I came across your project, I thought, I might buy this as a base for my router. But the thing was, it depends on the support of friendly elec for openwrt for example. And I don't want this dependency. So in terms of os support is your soc choice supported already, or will be supported out of the box?
What you are building is what i have been looking for , for some time, and in the end the rb5009 seemed to be as good as it gets at thr moment without custom x86 boxe. But here comes the interesting part, what will you do regarding the OS, openwrt is great i love it, yet , with support for opnsense, you will really set yourself as the defacto go to product for home labbers. I admire your work!
Please add the capabilities for it to be Rack mountable, so preferably 1U in Height and adaptable to be 19" if it's also small enough to be adapted to 10" that would be awesome but should not be a priority.
You could use the pcie from the cpu and build in support for a lte module on the motherboard, many companies like to use cell service as a backup in case their fiber gets cut. Could make this product easier to sell to a small business/medium business as this could be a requirement they look for in a router.
This is great! I have one feature I've been searching for on a board like this: M.2 M-Key (for NVMe SSD) with at least 2 lanes of at least gen3. With that feature, these could be excellent nodes in a distributed storage array (e.g. Ceph cluster). I believe units like that will eventually become popular with the same market segment of NAS, and certainly the market of homelab.
I am definitely going to follow in your journey. I would absolutely be interested if there was PoE support on at least 2 of the channels. Ideally all 4 but I know, a cost add thing for power. Would be a very interesting product if had that capability.
Thanks! I actually have a small notepad with features people ask about. So everything will be considered, if there's demand, obviously. Makes no sense to add stuff only a handful of people will end up using.
Actually having this device being able to be powered by PoE would make more sense for me. A firewall is not a switch, it should not deliver power to other devices. Making it provide PoE would require a BIG power supply, which, I think is completely missing the objective of having a small external power supply. But, having it powered by an existing PoE switch (like an Access Point or VoIP phone) would reduce power cord clutter and remove the need for an annoying power brick. Would it be possible to make the device power draw less than 25W and be able to fit within the 802.3at (PoE+) standard?
The Brits (& Ozzies etc) often pronounce it as r-oo-ter, in North America (Canada, USA) its often pronounced r-out-er. I switch between both as I am a Brit living in Canada (over 20 years). Thanks for this video and the journey to building the router. Amazing stuff! I will be buying one and if you go kickstarter, I will involve myself too!!
Hopefully if this project turns out to be a success you could make a future version with something like a Layerscape LX2122A CPU so that you can have a router with actual 10gb inspection throughput and not just NAT and filter throughput. Unfortunately, the price is around 300 which is a big step up. I do like that Ryzen V3000 idea though. I was just looking at those this last week myself. The performance on them is significantly higher than the other Intel CPUs you mentioned in the vid, and some of the V3000's should be capable of 10gb inspection throughput as well. I couldnt find a price for them unfortunately, but one company that sells a whole industrial PC (Israel company, with a few distributors in Europe) based on the chip sells the end product for anywhere between 950-1200 depending on the processor model.
I am really excited about this project and hope it succeeds. Unfortunately, it isn't the level of device I'm looking for as my current needs are something with 5gb IPS throughput, and I would like to get something with 10gb IPS throughput so I can keep using it as my main router/gateway into the future as two ISPs in my area plan to make 10gb plans available within the next 1-2 years. Currently, 5gb plans are $155/mo and 10gb plans are supposed to launch at $220. Gigabit plans have 3 providers in my area with prices ranging from $50-60 now 😁
Not sure how it may effect BOM, but maybe a N5095 or N100 will make more sense? They are considerably more compatible with OS and would also have a virtualization host option i creasing the market share you can target. (In addition to allowing a virtual FW, dns server and wifi mgmt software to run on the same hardware)
I think the youtube icon in the intro looks kinda off to the right, it seems to be in the center, but not in the visual center. I think webdesigners use a tool to find out the optical center instead of the technical, just some things I notice xD
I was thinking if it could be a bit like framework laptop with upgrade strategy and an protection against obsolete items. Split design on 2 pcb, one with main io and "stupids" electronics, then use a cheap memory socket to cpu board, which then have edge pins build into board. Then future could have different cpu boards, or in defects, only chance cpu board. Than also add a cpld/fpga in interface which can really protect against future changes in requirements, just ensure high speed / low latency interface - could extend interfaces also. 🤓
@@shanemshort massive i would not say, but really depends on how it is done, could however unlock future revenues also with upgrades or extensions. It could be what sets this product apart from the rest - just my two cents 🙂
Have been following your feeds. I fancied similar custom router but ended building the router from standard stuff; 19" rack server case, atx mobo, dual sfp card, dual 10gbit card and it runs pfsense. If I ever need to reconfigure the hardware its easy. Downside is power consumption, but it was quick to put together.
I am totally on the consumer side of this project, but loving every minute of it. While I am technically minded, I cannot wrap my head around PCB design. I will be following this project closely as I am very much so interested in this development process. In your questionnaire you asked if I would back you on Kickstarter. There was no room for additional information, so if would you allow me to do it here; I would probably not back you on Kickstarter as I have terrible experiences with KS. I lost a lot of money. And KS just says; "well, part of the game." Perhaps I backed the wrong people or projects, but still, I would be hesitant. Maybe other platforms? Looking forward to the next video!
Thanks for the feedback - I completely get what mean. I was more lucky with KS - every project I backed, shipped. But I didn't back many, so there's that. And welcome aboard!
Hello! You did a little mistake, you said that the RTC crystal oscillates at 32768 kHz instead of 32.768 kHz (32768 Hz). I really like your videos, I have just subscribed.
That's the one thing that's bugging me to no end. According to the datasheet, it should be possible to use it along with the 10 gig ports, but it's a bit,... ambiguous. I haven't gotten a straight answer from the distributor either, so I guess, I'll have to test it myself. It's no big deal, just takes a bit of time, because PHY chips usually have same footprints between 1gb and 2.5, so I can put one in place of the other.
I think the product will either live or die depending on what type of gigabit you chose. Its the most important part of the entire thing, and is basically table stakes!
Router -- In my 40 years as an engineer, trained and experienced working in the US Silicon Valley, I've never heard anyone refer to the WAN to LAN interface as a roo-ter. Your first pronounciation was connect. Rou- rhymes with cow, or wow! I've heard people from New York refer to major roads as routes, rhyming with root, and people from Atlanta call them a rut, but I've never heard anyone pronounce router any other way, no matter where in the country or the world I've been.
The thing is, it'll likely still require around 35W and I'd like to have some headroom, I'd need PoE++ (60W). I don't think there's enough potential customers out there with PoE++ capable gear out there.
Interesting, but I'd have to re-wire my entire place to make good use of such a router, wouldn't I? I currently use 1 Gbits...10 was not affordable, but I agree that it is the future!
Just coming into this a bit late I'm sure, but relating to the flash, would it be an idea to have user-replacable storage such as NVME or a SATA DOM? It would be great to see an internal USB port of some sort as well for USB-bootable OSs In the UK at least, we pronounce it 'root-er', but US calls it 'row-ter'
Not late at all, there's plenty of research and planinng yet to be done before this goes into an actual hardware design stage. We plan to have an M.2 port and another mPCIe one. It's not 100% for either, but they are being considered.
The only thing that i feel would be better would be 4 x 2.5 Gigabit ports instead of 4 x 1 Gigabit ports since there's been a trend in consumer devices to have 2.5 Gigabit ports built in
5V at 35W would be 7A, USB-C can do a maximum of 5A. For 35W you'll need USB-PD to get a higher voltage. At 12V and higher, you'll get 35W with any USB-C cable. PD won't add too much to the BOM these days
Less worried about the proc as long as it does the thing, or uses a ASIC to offload tasks too, but I have moved on to 10/25/50/100Gbps connectivity inside the house and would want to see 1 or 2 multi-gig (1,2.5,5,10g) ports for wan + one sfp+/sfp28 port.
I believe that's obvious but please use a power brick that has cables both to device AND wall outlet. Those laptop chargers with brick mounted on power plug absolutely suck and I can't imagine trying to plug such thing in a tight rack enclosure.
BTW: In case you have in mind to build a Ryzen Embedded-V router, I'm interested too. For me Intel starts to have to many issues with their silicons lately.
I'd personally prefer a DC jack rather than USB. USB is still a massive shitshow of interop issues with PD. A simple barrel jack is cheaper and easier in so many ways. I was going to suggest PoE, but that defeats the purpose of the 10G ports really. Another +1 for dual SFP+, 10G-T is power hungry and expensive.
This seems to be a recurring request (dual SFP+). I've actually started making measurements of the planned PCB to check how it fits. Will make a video with all the updates that came from the comments.
Hello! Very interesting process and I wish you best of luck with it! :) I`ve read through below comments and I didn`t see anyone mentioning the HW acceleration for PPPoE WAN connection. Without it, your CPU will be totally under the water and even cheep alternatives (like the ones below 100$) will easily surpass the gigabit PPPoE performance, not to mention the above ones, 2.5GGbps or even 10Gbps. Ubiquiti had this issue and they worked on it for more than a year to optimize the OS and distribute the encryption on multiple CPU cores. Why I`m mentioning this? -> Most of EU countries still have PPPoE fiber connection through GPON and now some of them, like Romania, are transitioning to multiple gigabit (2.5G/10G through XGSPON). :) If your router will be capable of 1Gbps PPPoE with IDS/IPS active or 10Gbps without IPS/IDS you can consider me as client. :)
Have you considered a cheaper arm CPU and a switch chip/ASIC? The processor could be just handling 2x1G or 2x2.5G and having the second port internally connect to a dedicated switch chip. Ie: 1G/2.5G WAN port -- processor -- internal 1G/2.5G connection -- switch chip -- multiple LAN ports It would add some development complexity, but would provide a potential upgrade pathway with better switch chips and processors for future SKUs/iterations. I havent spent much time looking at prices, but it *might* also reduce the BOM. I'm keen to hear your thoughts on this. Australian, English is my first and only language. In router, we say "ow" like "how"
I'm also curious why not just 2 SFP+ ports, since there are RJ45 transceivers available. This would be functionally still allowing RJ45, but also enable dual optical.
Dont know which video to post this one.... Came across this product just now while I was looking up VyOS on ARM. From the specs it looks like your router you want to build. Using the 1088 instead of the 1064 SoC CPU. The Traverse Technologies Ten64; mind you its got some better specs (higher RAM, and I think it was more cores) but thought I would bring it up to you just incase you havent seen it yet.
@@tomazzamanRoger that. 😊 Actually had an idea this morning for a video for you… if you got this far in and then found a product similar to what you are designing what to do / how to decide if its worth while continuing
I'm wondering about the performance of the router. Love the project btw, I think it's awesome to tackle a project like this and good luck! Only thing I'm thinking about is how these NIC's will compare to Intel NIC's. I have run multiple Pfsense boxes, always had performance bottlenecks until is started using Intel NIC's. Can I expect this router to have similar performance? I'm really unfamiliar with using Arm and Marvell to build a router.
Once the development kits arrive, I’ll test the CPUs thoroughly to make sure the throughput is as advertised. Of course I’ll report my findings here on the channel!
Interesting, you claim that you can get ~1Gbps IPS/IDS how do you calculate that? I have estimated that the LS1046 should manage a bit more than 10Gbps NAT with large packets. And i assume that the processing is then 10 times heavier but do you have any data from another CPU showing its performance?
Hi Tomaz. When you predict to be available? I tried a Qotom C3758R with 4x10GbE SFP+ and 5xI226-V and unfortunately all I226-V NICs are triggering Hardware error from APEI Generic Hardware Error Source: 0, AER: aer_status: 0x00002000, aer_mask: 0x00002000, AER: aer_layer=Transaction Layer, aer_agent=Receiver ID and the driver is crashing. Sometimes is just triggering the Hardware error and sometimes is also crashing. I have no clue if is PCIe issue in CPU or something else. I can't believe that all 5 NICs are throwing the same error. These errors are triggered on iperf3 --bidir. I couldn't manage to trigger them if I'm doing only one-way traffic. I really need a good device and your looks very promising.
After watching this video and the previous one, I still don't understand the choice of an ARM processor. The most likely candidates for firewall softwares on that new device would be pfSense, OPNSense and VyOS and none of them support the ARM architecture yet. I don't think that hoping for those product to release an ARM version by the time you put your hardware on the market is realistic. If you don't have a killer app for your hardware, it will not sell. Very few people will be willing to pay more than 400 USD for a hardware just to be able to run OpenWRT on it. What about the Intel Atom C1110?
Please Please Please make both 10G ports SFP+. You can always plug in an RJ45 SFP. But if you don’t have 2 SFP ports you can never have Fiber in and Fiber out. This is my need.
Is it possible to upgrade the 1 Gbit Ports to 2.5GBASE-T and PoE? There isn't much point in having a 10Gbit Internet or home server connection if all other connections only add up to 4 Gbit in total. And 2.5GBASE-T together with PoE is realy great for connecting a couple of Wifi-6 access points across your home or office.
2:15 with this cpu, this will be a software-router? No ASIC? A ddos attack with a lot of PPS will consume all cpus with IRQ? This remember me a Mikrotik 😅. Thanks!
LS1046 (our CPU of choice) has something called a DPAA, meaning it has dedicated hardware for a lot of network functions. So it has best of both worlds, for an acceptable price: 4 general purpose ARM cores and the specialized DPAA.
don't use 5v as input, please support proper usb-pd so that you can use any power supply, otherwise it makes 0 sense and it would be better have a barrel jack. 15w for usb-c power supplies are meant to be delivered over 9v and not 5v for most power supplies
@@tomazzaman I just don’t want to see another raspberry pi situation where you need a specific power supply capable of 3A+ over 5v while you can buy proper usb-pd of 35w for 20€
I wish that you could add 4 x 2.5Gbe ports instead of 1Gbe. But i guess yoiu cannot, so at least try to make the 10Gbe port, backwards compatible with 5/2,5/1 Gbit speeds. I wish you the 2024 bring your dream alive.
DPDK supports LS1046A. Will this hardware be validated for DPDK? Will you support upgradable RAM and storage? If I can run Linux with DPDK and upgrade to 64GB, and upgrade storage (m.2 pcie or at least mSATA) this would be great for my software project.
Some tips from a former embedded developer. The PHYs will also need a clock. But if they use the same frequency (e.g. 125 MHz) you can reuse the same oscillator and use clock buffers or daisy chain the clock signal. Saves you another chip. Also there are oscillator chips with multiple output clocks and you can program every output clock before hand.
Another important thing are LEDs, they are useful for debugging but also have their use during production and can show if some software component is running or not. Just make holes in the case and add walls to block light bleeding. People will find many uses for the LEDs.
Former? What made you quit/leave?
Thanks for the suggestions regarding the clocks, current candidate is 6V49205BNLGI8 (and 9DML0441AKILF for the buffer).
You have just made me think about the LEDs though, I personally hate them anywhere else but the ethernet ports (so that I can see activity). The initial plan was to have one LED that blinks while booting up and then remains lit during normal operation. However, putting some extra on PCB for debugging purposes makes a lot of sense. Thanks!
@@tomazzamanNothing dramatic, I left the company and started my master studies. In a few months I have finished and will return to the job.
I definitely agree with adding leds. I recently finished an embedded design where I was required to show all the power, status, and error information with one RGB LED. It took a lot of development time to figure out how to squeeze all that information through one LED when it would have been dead simple to have separate LEDs for each function.
If I was doing that design again, I would use separate power, status/activity, and error LEDs. Unless PCB space is *the* limiting factor, I would at least add pads for indicator LEDs. If you don't use them there's no wasted cost, but if you need them later you can do light pipes to the case which was easier then I thought it would be
I like dual color red/green leds for status and error since you can do red/green/both (amber)/black (off) flashing for intuitive severity feedback. People can guess that fast flashing red/black is clearly worse than slow green/amber. Just my two cents having flashbacks (pardon the pun) to working on UI with LEDs
Personally I am still a simp for ye-old 7-segment displays. It just makes it so much easier than just using a few random LEDs.
@@tomazzaman in case of LEDs you may want some buttons, too. you can turn the aditional LEDs of, and leave only power LED in a dimmed state, on button press, you activate the rest of the leds to display certain status info, like sfp status an online indicator, an vpn indicator, or anything else a user can imagine.
maybe take a look at the Speedport Smart 4 from Deutsche Telekom and use an OLED Display. There are multiple use cases for an display. heartbeat, actual amount of traffic, load, number of active (WiFi) clients, IP Adresses, CPU Load
with an display you are not limited by the amount of leds and also can utilize it as an led replacement.
IMHO the deciding factor for buying this router would be mainline kernel support. The number of SBC style boards I've purchased over the years which lasted < 2 years because I wasn't able to update the software on them anymore because the vendor used the chip manufacturers custom kernels to build forked variants of whatever os/distro.
Agreed, would help with OpenWRT long term use.
Not a huge fan of the external power brick, but it makes a lot of sense, heat, space, etc.
Also if you were to go internal people would want redundancy, it’s a losing battle.
That said, as another commenter said proper implementation of USB C PD would be nice!
If you could add in POE Base power via one of the ethernet ports would be amazing
Devices this small always have external power bricks though, and they are extremely common for routers and other gateway type devices. I see nothing wrong with it.
Though as @lachlanstone282 said, it would be very cool if next to the DC input port you had an RJ45 port with the communication lines disconnected and only use the lines that poe+ power comes in on. It would be really nice to use a port on the main network switch that is either poe+ or poe++ and power the router like that for those of us who had the capability. But in the end, that is just a cool idea that would likely be used by few people and add to the cost.
Yes, those PoE-powered switches are great! I agree with your points, though. I also agree with what you are saying about power bricks; the exception is for rack-mounted devices. @@LordSaliss
@@LordSaliss just switch 1 rj45 port to rj45 w/ PoE input ;)
The power brick makes it easier to launch a product else you're going to go through a cert route with added costs for 110/230VAC EMI and stuff and now you just say use a usb-c power brick.
I think what you're doing is really great. I only just found this channel and have started going through your router designing videos. Really enjoying it so far.
BUT in my opinion, not having dual SFP+ is a total deal breaker; For me and probably for a lot of other people too. Especially those using fibre networks.
So considering the CPU's limitations, it makes much more sense to have dual SFP+ instead of 1xSFP+ and 1xRJ45 for the 10Gb ports. That way SFP+ fibre can be used for both WAN and LAN. Then if the user does need RJ45 they can just use an RJ45 Transeiver. I think not going dual SFP+ will reduce the market that your router will appeal to. I also think that with dual SFP+, your router will have the potential to be marketed as a professional router to the SME (Small and medium enterprise/business) sector, not just homelabbers. And if priced correctly could be extremely competitive in the small-medium business market. Even against Cisco and Ubiquity's offerings. You could have a router that is damn near perfect in my eyes.
Thanks for the suggestion, it seems more people prefer 2x SFP+ as opposed to 1x 10Gb RJ-45. I'm seriously considering it.
@@tomazzaman consider the ryzen SoC, you get 2x10Gb/s interfaces for embedded and a vega GPU but this is with their gen 1 ryzen embedded. Might have newer variants as mine is a bit old. This lets you easily have SFP+ as with SFP+ you can reduce some of the costs associated with PHY ethernet, however this puts the burden on customer instead for ethernet. However SFP+ direct is cheaper than 10Gb/s ethernet in short distances with ensured cable quality from my experience. After a long time running ethernet sometimes they tend to drop in speed which is annoying. So with 2x SFP+ one may put GPON opticla on one, and on the other SFP+ direct to a SFP+ switch. In my previous setup i had a 10Gb/s capable switch with 36 TILE cores with 2x SFP+, one of it to a SFP+ switch which i combine to ports to another switch before breaking it out to 24 ports. On the SFP+ switch itself i also connect file servers.
I can help you find better SoC options for your project than what you are currently using. However check out ryzen embedded SoCs first as well as other ARM options. Have you also tried snapdragon given they have good OS support? intel atom is very slow in comparison to rockchip and better options.
@@tomazzamanIgnore 1/2.5gb ports and go 10gbe all the way. You give users the option of using different connections if you split up the ports evenly (3xSFP+/3×rj45).
A router with just two SFP+ ports and nothing else would be kick ass.
@@tomazzaman Combo port for 10G, best of both worlds. User choose what media (RJ45/SFP+) they need, like some enterprise switch.
In my experience, sfp+ is more flexible and reliable than the rj45 variant. So better use 2x sfp+ instead of 1 sfp+ and 1 10gbit rj45. When you really want rj45, you can always add a sfp+ rj45 module.
I would really like a dual SFP+ variant as we always use sfp+ with fiber in almost all our builds, we never use the rj45 variant
Agreed. This is incredibly important to make sure all use cases are covered. As far as I'm concerened, not having dual SFP+ is a total deal breaker for me.
Personally I think having dual SFP+ should just be the default and only variant. If users need RJ45, they can just use adaptor transeivers.
Hm, I'll give it some thought. Dual SFP+ would actually make the device cheaper.
Problem is that in Europe the 10g port on the modem for FTTH is a 10g baset sadly rather than sfp+
My thoughts to. SFP+ modules for fibre or copper are fairly cheap now (sub $20) and come in both 1 and 10G options. Doing this lowers the cost of the unit while allowing the user to pick what they need. Would definitely be a requirement for me.
Agree with this too. 10G over copper tends to be more of a backward-compatibility use case, and it can still be supported as such with a 10G-BASE-T SFP+ module.
8:19 I think of it this way, I say out, not oot.... So R-out-er, not R-oot-er ;)
Loving these deep dive videos as you work all this out. Looking forward to the rest of your journey.
For the uart , please consider using a standard RJ45 rolover type console port in addition to the type-c console port. ive had terrible luck with built in uart -> usb whereas with rj45 i have a pile of adapters that i know work every time.
Well done. Great visualization of the block diagram 💪.
I'm excited to see this project develop. 3 things I'd like to see, rackmountable with pigtails, 2x sfp+ rather than 1xsfp+ and 1x10gb, power input on the back. Good luck!
Thanks, it seems I'll have to seriously reconsider the 1x RJ-45 as there seems to be more and more people asking for an additional SFP+
Thanks!
@@tomazzaman I like the idea of 1xRJ-45. The SFP+ to RJ45 transceivers generate a hell of a lot of heat. Also here in the US we get 2G fiber but the ONTs still connect via RJ45. You have to cover a broad audience. Not everyone will connect direct to SFP+ and not everyone will connect at 10G. Most of us are just getting into 2.5G. Also Rackmount option!
@@harihoudini Ubiquiti dont want to support 2.5G SFP modules. Their fans are fanatic and i wonder why people still want to pay their prices for bad options. Other brands already support it. The support is more down to software though. Since datacenters dont use ubiquiti they have no one to scream them into action meanwhile if mikrotik doesnt support a feature many are going to scream at them as businesses and datacenters do use them.
Some ARM SBCs with 16GB of ram for $150 faster than intel atoms are coming with 2x2.5Gb/s ethernet.
@@tomazzaman1 x 10Gb RJ45 and 1 x SFP+ are perfect. Wouldn’t even need the 4 x 1Gb RJ45. Just need a router with 1 x 10 Gb RJ45 to connect to the Fiber modem and 1 x SFP+ to connect to a 10 Gb Switch. And a rack mount option. That is all I am looking for. Didn’t find anything reasonably priced so far and have high hopes in you project. Anything extra only makes it more costly and I hope it stays below 600-700 €/$.
@@tomazzaman Count me in the "2xsfp+ = interesting, 1xsfp+ and 1 RJ45 = no thanks" camp.. as may other said from sfp+ to rj45 is a fairly cheap device (addon option perhaps, when selling?)...
Its really nice to see the way you think, very interesting. It s great to see finally someone wanting to create a quality router. Great job
I’m very interested in this project. I speak US English and while it’s commonly pronounced both ways I prefer router, not “rooter”. I prefer an external power brick, as long as you use something standard like usb-c PD or a 12V 5/2.5mm barrel jack. This benefits replacement of a bad power supply, minimizes package size & heat dissipation, and greatly simplifies regulatory compliance (CE or UL).
13min master class! very good! im in this "game" since 20 years now and most people dont know about this basics.
Thanks for the kinds words!
Why not 2.5GBE instead of gigabit Ethernet.
2.5GBASE-T is a hell of a standard ...
@@deardiso9193why is that?
The CPU only has gigabit and 10G
@@benchy5769It looks like the CPU supports one 2.5GbE port based on the CPU block diagram at 9:42
2.5GBe would be highly appreciated. I wouldn't want gigabit if I'm upgrading right now
I was some week ago learning about make my own router. It's awesome see your plans to build. Hope it works and have international shipping 🙌
Thank you. If we succeed, we'll ship worldwide, of course!
You should definitely provide some way to connect 12v power through a DC barrel jack or possibly 48v if you want to support POE. Pretty much all networking gear runs on those voltages / connectors, and it's much easier to supply power at higher voltages. If you have USB-C power input you would need the switching electronics anyway because USB-C PD runs at much higher voltages (usually 20v, but it's negotiated by the charger and device) in order to deliver 10s of watts of power over relatively thin wires. Buck converters are relatively small and inexpensive these days, so you can internally downconvert it to 5 or 3.3v as necessary.
Don't know when was the last time I heard someone say usmerjevalnik 😄
Im curious to see the OS implementation plans, it seems like opnsense has some unofficial support for ARM64 but it will likely need the latest branch of BSD to be ported to your platform which sounds costly. Hardware plans are awesome and im excited to see where it goes!
Thanks! I'm currently in talks with another vendor of telco gear that can lend me some ARM-based PCBs for me to try and test how difficult it is to port opnsense to it. Will report here, of course :)
@@tomazzaman was googling and came to this: "OPNsense 23 for aarch64" by yrzr on 2023-07-13 *screenshot says a ARM Cortex R72
Given the security implications of being a router in a network, did you consider:
- adding UEFI secure boot with a PKI where you can give out certificates for "official" distro maintainers for your router that you trust, for end-users that still want something powerful but don't want to be a "maintainer of a router" so to say?
- are you planning on using any hardening on the OS side like SELinux?
- for example i'm running a raspberrypi with fedora-iot which uses rpm-ostree, systemd with podman rootless containers, etc. I'm currently looking into SELinux to harden it even more. It's nice because with 'systemd-analyze security' you can also easily analyse a lot of security related knobs for your containers/services.
Good luck and thanks for all your awesome contributions to us (videos, knowledge, hardware, etc.)
Have you considered going with dual sfp+ ? SInce RJ45 transceivers would allow for dual function. It would also be nice if one of the 1GBE ports was a SFP port, since that would allow one to not have to use a ISP CPE. USB-C PD would also be very nice.
SFP/SFP+ RJ45 can get hot and be prone to failure in some systems
Like @omegatotal said, RJ-45 SFP+ modules get VERY hot (I have one in my switch, can attest to that), so I'd rather have the flexibility. Putting two side by side would be a risky move. Don't get me wrong though, it would be cheaper to manufacture as we'd get of one PHY chip which costs around $30 on its own, but I'd rather not force potential customers to buy more stuff to get it working.
I and almost all installs we do use fiber either OS2 or OM4 we would rather have 2x sfp+ ports seen as we run fiber between all other gear and fiber comes in in fiber format so the rj45 would be un used and that way useless or were are force ro run a sfp+ rj45 on the switch side
A huge thing for the potential cost is the longevity of the device. With an arm SoC you end up relying on someone maintaining a specialised kernel. An embedded x86 processor would give buyers the knowledge that the device can be repurposed in future and isn’t locked down (could run pfsense etc). I’d also suggest 2x sfp+ and 12v barrel jack. The 4 gigabit ports seem like a waste IMO, anyone forking out for this will likely have a core switch behind it, it may be better to offer a PCIe slot so buyers can choose their own expansion, if they want 4 gigabit it 2.5g or 10g it’s up to them.
Thanks for the feedback, apart from the ARM SoC I'm addressing the rest in my pricing video. We've indeed moved from 1x4Gb to 3x2.5Gb (because we can do that with the same CPU). Also 2x SFP+! :)
x86 unfrotunately brings too many problems for a small startup to solve. And ARM CPUs can just as easily be repurposed. Yes, the support is smaller, but they are gaining in popularity, so it'll get better with time.
@@tomazzaman that’s fair enough, part of engineering!
I really like the progress you are doing. I think that it would be nice to have redundant power inputs if it would not ad that much to the cost. It might be cool to be able to plug a PD capable powerbank to one of the power inputs as a UPS in that case.
USB C power delivery would be great to have instead of a brick with specific voltages and plug
For total of 35W (11:43) of power a 5V over single USB Type C input is not enought. Even high-current cables max out at 5A (5A*5V=25W, good luck with chargers which even support that, most likely 5A could only be used at 20V), standart current cables max out at 3A. I would recomment using 12V or 15V USB PD (power delivery) input for 36W (12V 3A) - 45-75W (15V at 3A or 5A if you are lucky)
Neat idea. Hope you gus cover the software bootstrap part as well. Thank you.
Also, with this size, a caseless version would be acceptable as a cas can be 3D printed!
I suggest adding optional brackets to the case for rack mounting
Yep, planned.
Can you make a router that is upgradable.
Upgrade radio
Upgrade RAM
Upgrade the main board.
So this will make it better to not sell the entire router everytime one needs to upgrade, the router ‘s robust amazing aluminium can be kept as it is.
I think the bottom line of the cpu choice is, that the bootloader is open source, or follows certain standards - UEFI support would be great!
Odlično! This is the most interesting project these days on the Internet. Just please, make it 2.5 GbE instead of 1GbE.
Btw what is your background?
2.5GbE is only possible on one port because of the CPU limitations. I mean, I'd love to choose better components, but then the device will become much more expensive and inaccessible to some.
My dad had an aluminium foundry, so I helped him a lot growing up. Because of that I signed for mechanical engineering middle school, but I partied too much so my grades weren't that good. Went to Biotechnical faculty to become a wood technology engineer, but didn't finish as I got kids and fell in love with software development. In 2012 I started a startup called codeable.io (with a partner), left last year. Now I do hardware. Or, I'm trying to :)
@@tomazzaman it make sense, given the CPU limitations. On the other hands, we are getting cheaper 5x2.5GbE + 2 10GbE (unmanaged) switches that are super affordable. I would gladly pay more for something with more capabilities. Maybe even PCI-E expansion slot as on Zima board, so one can choose the expansion. But I guess that's even more complex and expensive on the CPU side. Can't wait for the next video and some rough estimation on the pricing.
Cool that you are able to pursue what interest you. Wish you all the luck with the project(s).
My dad has a car/auto-service, so I started with modifying cars, then racing, now sim racing online. I studied computer science on a mathematical faculty (PMF) so I naturally went into software. But I'm so interested in hardware, IoT, smart homes and all of DIY stuff. That's why I find this channel super cool. Kudos!
I would like to know how it would perform in comparison to the Banana Pi BPI-R4.
You know what? It seems this is a recurring comment, so I'll order one of the Bananas and test to compare! 🙌
@@tomazzaman before I came across your project, I thought, I might buy this as a base for my router. But the thing was, it depends on the support of friendly elec for openwrt for example. And I don't want this dependency.
So in terms of os support is your soc choice supported already, or will be supported out of the box?
What you are building is what i have been looking for , for some time, and in the end the rb5009 seemed to be as good as it gets at thr moment without custom x86 boxe.
But here comes the interesting part, what will you do regarding the OS, openwrt is great i love it, yet , with support for opnsense, you will really set yourself as the defacto go to product for home labbers.
I admire your work!
Odlično Tomaž! Veliko uspeha še naprej!
Please add the capabilities for it to be Rack mountable, so preferably 1U in Height and adaptable to be 19" if it's also small enough to be adapted to 10" that would be awesome but should not be a priority.
Yup, planned. Haven't mentioned it anywhere, but we'll get there! Thanks!
You could use the pcie from the cpu and build in support for a lte module on the motherboard, many companies like to use cell service as a backup in case their fiber gets cut. Could make this product easier to sell to a small business/medium business as this could be a requirement they look for in a router.
Yep, a PCIe port is planned and we're also brainstorming ideas as to what it should support. Thanks for the input!
This is great! I have one feature I've been searching for on a board like this: M.2 M-Key (for NVMe SSD) with at least 2 lanes of at least gen3. With that feature, these could be excellent nodes in a distributed storage array (e.g. Ceph cluster). I believe units like that will eventually become popular with the same market segment of NAS, and certainly the market of homelab.
I am definitely going to follow in your journey. I would absolutely be interested if there was PoE support on at least 2 of the channels. Ideally all 4 but I know, a cost add thing for power. Would be a very interesting product if had that capability.
Thanks! I actually have a small notepad with features people ask about. So everything will be considered, if there's demand, obviously. Makes no sense to add stuff only a handful of people will end up using.
Actually having this device being able to be powered by PoE would make more sense for me. A firewall is not a switch, it should not deliver power to other devices. Making it provide PoE would require a BIG power supply, which, I think is completely missing the objective of having a small external power supply.
But, having it powered by an existing PoE switch (like an Access Point or VoIP phone) would reduce power cord clutter and remove the need for an annoying power brick. Would it be possible to make the device power draw less than 25W and be able to fit within the 802.3at (PoE+) standard?
The Brits (& Ozzies etc) often pronounce it as r-oo-ter, in North America (Canada, USA) its often pronounced r-out-er. I switch between both as I am a Brit living in Canada (over 20 years). Thanks for this video and the journey to building the router. Amazing stuff! I will be buying one and if you go kickstarter, I will involve myself too!!
Thank you! appreciate the support!
"A good plan is half the Job done" Nope 😛
“Plans are worthless, but planning is everything,” - Eisenhower
Hopefully if this project turns out to be a success you could make a future version with something like a Layerscape LX2122A CPU so that you can have a router with actual 10gb inspection throughput and not just NAT and filter throughput. Unfortunately, the price is around 300 which is a big step up. I do like that Ryzen V3000 idea though. I was just looking at those this last week myself. The performance on them is significantly higher than the other Intel CPUs you mentioned in the vid, and some of the V3000's should be capable of 10gb inspection throughput as well. I couldnt find a price for them unfortunately, but one company that sells a whole industrial PC (Israel company, with a few distributors in Europe) based on the chip sells the end product for anywhere between 950-1200 depending on the processor model.
I am really excited about this project and hope it succeeds. Unfortunately, it isn't the level of device I'm looking for as my current needs are something with 5gb IPS throughput, and I would like to get something with 10gb IPS throughput so I can keep using it as my main router/gateway into the future as two ISPs in my area plan to make 10gb plans available within the next 1-2 years. Currently, 5gb plans are $155/mo and 10gb plans are supposed to launch at $220. Gigabit plans have 3 providers in my area with prices ranging from $50-60 now 😁
Thanks for the feedback!
Not sure how it may effect BOM, but maybe a N5095 or N100 will make more sense?
They are considerably more compatible with OS and would also have a virtualization host option i creasing the market share you can target. (In addition to allowing a virtual FW, dns server and wifi mgmt software to run on the same hardware)
I think the youtube icon in the intro looks kinda off to the right, it seems to be in the center, but not in the visual center. I think webdesigners use a tool to find out the optical center instead of the technical, just some things I notice xD
I was thinking if it could be a bit like framework laptop with upgrade strategy and an protection against obsolete items.
Split design on 2 pcb, one with main io and "stupids" electronics, then use a cheap memory socket to cpu board, which then have edge pins build into board. Then future could have different cpu boards, or in defects, only chance cpu board. Than also add a cpld/fpga in interface which can really protect against future changes in requirements, just ensure high speed / low latency interface - could extend interfaces also. 🤓
that strikes me as a massive amount of added cost
@@shanemshort massive i would not say, but really depends on how it is done, could however unlock future revenues also with upgrades or extensions. It could be what sets this product apart from the rest - just my two cents 🙂
Have been following your feeds. I fancied similar custom router but ended building the router from standard stuff; 19" rack server case, atx mobo, dual sfp card, dual 10gbit card and it runs pfsense. If I ever need to reconfigure the hardware its easy. Downside is power consumption, but it was quick to put together.
I am totally on the consumer side of this project, but loving every minute of it. While I am technically minded, I cannot wrap my head around PCB design. I will be following this project closely as I am very much so interested in this development process. In your questionnaire you asked if I would back you on Kickstarter. There was no room for additional information, so if would you allow me to do it here; I would probably not back you on Kickstarter as I have terrible experiences with KS. I lost a lot of money. And KS just says; "well, part of the game." Perhaps I backed the wrong people or projects, but still, I would be hesitant. Maybe other platforms? Looking forward to the next video!
Thanks for the feedback - I completely get what mean. I was more lucky with KS - every project I backed, shipped. But I didn't back many, so there's that.
And welcome aboard!
Hello!
You did a little mistake, you said that the RTC crystal oscillates at 32768 kHz instead of 32.768 kHz (32768 Hz).
I really like your videos, I have just subscribed.
Thanks for the correction! And welcome aboard!
That magnetics "shielding" against EMI won't stop any external interference.
I would run with a eMMC instead of a NAND flash, and possibly skip the NOR. Should be reliable enough.
Sorry didn't realize this video was 8 months old
that 2.5G port from that cpu ought to be used too
That's the one thing that's bugging me to no end. According to the datasheet, it should be possible to use it along with the 10 gig ports, but it's a bit,... ambiguous. I haven't gotten a straight answer from the distributor either, so I guess, I'll have to test it myself. It's no big deal, just takes a bit of time, because PHY chips usually have same footprints between 1gb and 2.5, so I can put one in place of the other.
I think the product will either live or die depending on what type of gigabit you chose. Its the most important part of the entire thing, and is basically table stakes!
Im german, and we also call it Router , but more spoken like you would pronounce Rooter / or "Ruuter"
in spain i think we mostly pronounce it /ˈruteɾ/ (written following the IPA)
Router -- In my 40 years as an engineer, trained and experienced working in the US Silicon Valley, I've never heard anyone refer to the WAN to LAN interface as a roo-ter. Your first pronounciation was connect. Rou- rhymes with cow, or wow!
I've heard people from New York refer to major roads as routes, rhyming with root, and people from Atlanta call them a rut, but I've never heard anyone pronounce router any other way, no matter where in the country or the world I've been.
its an accent thing, many languages that are not english pronounce it "rooter" so maybe that's where his pronounciation coming from
Really great Project! Since there are no high power requirements, what about POE to power the board?
The thing is, it'll likely still require around 35W and I'd like to have some headroom, I'd need PoE++ (60W). I don't think there's enough potential customers out there with PoE++ capable gear out there.
Interesting, but I'd have to re-wire my entire place to make good use of such a router, wouldn't I? I currently use 1 Gbits...10 was not affordable, but I agree that it is the future!
Just coming into this a bit late I'm sure, but relating to the flash, would it be an idea to have user-replacable storage such as NVME or a SATA DOM? It would be great to see an internal USB port of some sort as well for USB-bootable OSs
In the UK at least, we pronounce it 'root-er', but US calls it 'row-ter'
Not late at all, there's plenty of research and planinng yet to be done before this goes into an actual hardware design stage. We plan to have an M.2 port and another mPCIe one. It's not 100% for either, but they are being considered.
Cool stuff
The only thing that i feel would be better would be 4 x 2.5 Gigabit ports instead of 4 x 1 Gigabit ports since there's been a trend in consumer devices to have 2.5 Gigabit ports built in
You might want to check my latest video 🤓 hint: 2.5Gb ports. 💪
@ 8:08 - it's ROOter, as a ROWWter is used to cut wood :D
I think it doesn't matter how people pronounce words. As long as your ideas are understood is all that matters.
I've had a Gigabit home internet connection for almost 5 years
5V at 35W would be 7A, USB-C can do a maximum of 5A. For 35W you'll need USB-PD to get a higher voltage. At 12V and higher, you'll get 35W with any USB-C cable. PD won't add too much to the BOM these days
Yep. PD is planned. Because of the feedback in the comments - a lot of people would love to have it included.
Hardcore DIY!
In Brazil we call ROTEADOR
As an engineer, I'would be thrilled, if I can see an x86 PCB design! :) No pressure however! :D
Love the project, but you have the Minisforum MS-01 to beat in price/performance/features now.
Challenge accepted. :)
The market already has plenty of options for 1+ 10Gbps LAN/WAN ports and 4+ 1Gbps LAN ports. Would be nice to have 4x 2.5Gbps LAN ports instead.
Then you'll be happy to learn we've moved on to 2xSFP+ and 3x2.5Gb RJ-45! th-cam.com/video/GKj5P4goBS4/w-d-xo.html
Wow! What a find is this video! I Have already subscribe. And want to suggest a video about MII end PHY. Thanks for this content!
Welcome aboard! Video about PHY already in the works! :)
I pronounce router same as you. In Oz a rooter is something different 🤣
Less worried about the proc as long as it does the thing, or uses a ASIC to offload tasks too, but I have moved on to 10/25/50/100Gbps connectivity inside the house and would want to see 1 or 2 multi-gig (1,2.5,5,10g) ports for wan + one sfp+/sfp28 port.
I believe that's obvious but please use a power brick that has cables both to device AND wall outlet. Those laptop chargers with brick mounted on power plug absolutely suck and I can't imagine trying to plug such thing in a tight rack enclosure.
What I'd love is to have a tiny GaN wall-wart and a USB-C cable. We'll revisit this topic in a dedicated video.
BTW: In case you have in mind to build a Ryzen Embedded-V router, I'm interested too. For me Intel starts to have to many issues with their silicons lately.
I'd personally prefer a DC jack rather than USB. USB is still a massive shitshow of interop issues with PD. A simple barrel jack is cheaper and easier in so many ways. I was going to suggest PoE, but that defeats the purpose of the 10G ports really.
Another +1 for dual SFP+, 10G-T is power hungry and expensive.
This seems to be a recurring request (dual SFP+). I've actually started making measurements of the planned PCB to check how it fits. Will make a video with all the updates that came from the comments.
Hello!
Very interesting process and I wish you best of luck with it! :)
I`ve read through below comments and I didn`t see anyone mentioning the HW acceleration for PPPoE WAN connection.
Without it, your CPU will be totally under the water and even cheep alternatives (like the ones below 100$) will easily surpass the gigabit PPPoE performance, not to mention the above ones, 2.5GGbps or even 10Gbps.
Ubiquiti had this issue and they worked on it for more than a year to optimize the OS and distribute the encryption on multiple CPU cores.
Why I`m mentioning this? -> Most of EU countries still have PPPoE fiber connection through GPON and now some of them, like Romania, are transitioning to multiple gigabit (2.5G/10G through XGSPON). :)
If your router will be capable of 1Gbps PPPoE with IDS/IPS active or 10Gbps without IPS/IDS you can consider me as client. :)
Thanks for pointing that out, I was not aware of it! Will test the development board thoroughly to make sure PPPoE works as it should.
GIVE ME THE CHIPS!!! I WANT SOUR CREAM AND ONION!! I WANT BBQ SOZ!!! I WANT SALT AND VINEGAR!!!! I NEED CHIPS!!
Have you considered a cheaper arm CPU and a switch chip/ASIC? The processor could be just handling 2x1G or 2x2.5G and having the second port internally connect to a dedicated switch chip.
Ie:
1G/2.5G WAN port -- processor -- internal 1G/2.5G connection -- switch chip -- multiple LAN ports
It would add some development complexity, but would provide a potential upgrade pathway with better switch chips and processors for future SKUs/iterations.
I havent spent much time looking at prices, but it *might* also reduce the BOM.
I'm keen to hear your thoughts on this.
Australian, English is my first and only language. In router, we say "ow" like "how"
I'm also curious why not just 2 SFP+ ports, since there are RJ45 transceivers available. This would be functionally still allowing RJ45, but also enable dual optical.
There will be 2 SFP+ ports, releasing the update in today's video! (Not out yet)
Usb c power would be excellent
Yep, already planned for the first version!
Rowter pronunciation is American rooter pronunciation is English. Great videos.
Yeah, it seems there's quite a dilemma on how to properly pronounce it. And I love it! 😂
@@tomazzaman in British English, at least, a router (rowter) is a tool for making channels in wood, a router (rooter) is the device you're making.
That's a lot of RAM/ROM, especially compared to the specs of the Fritz!Box. For what do you need that much of RAM/ROM?
I pronounce router as router ;D
Interesting videos, though! I like to see deep dives of how people think when designing products. A sub well earned.
Thank you! There will be much more of that!
Are you going to add a coincell battery for the RTC if you disconnect power?
Dont know which video to post this one.... Came across this product just now while I was looking up VyOS on ARM. From the specs it looks like your router you want to build. Using the 1088 instead of the 1064 SoC CPU. The Traverse Technologies Ten64; mind you its got some better specs (higher RAM, and I think it was more cores) but thought I would bring it up to you just incase you havent seen it yet.
I've seen it yes, but 1088 is much more capable, and also much more expensive chip.
@@tomazzamanRoger that. 😊
Actually had an idea this morning for a video for you… if you got this far in and then found a product similar to what you are designing what to do / how to decide if its worth while continuing
I'm wondering about the performance of the router. Love the project btw, I think it's awesome to tackle a project like this and good luck!
Only thing I'm thinking about is how these NIC's will compare to Intel NIC's. I have run multiple Pfsense boxes, always had performance bottlenecks until is started using Intel NIC's.
Can I expect this router to have similar performance? I'm really unfamiliar with using Arm and Marvell to build a router.
Once the development kits arrive, I’ll test the CPUs thoroughly to make sure the throughput is as advertised. Of course I’ll report my findings here on the channel!
it should support virtualization.
Tbh I’d prefer 2 SFP+ ports. SFP to 10Gb Ethernet adapters are fairly inexpensive and I’d rather have the flexibility of SFP.
Your timing is spot on. Currently making an update video in which we do that change (and a couple of others)
yes router the first one
I would aim for 2.5gb IDS/IPS for us market, 2.5gb and 5gb fiber is becoming widely available
Interesting, you claim that you can get ~1Gbps IPS/IDS how do you calculate that? I have estimated that the LS1046 should manage a bit more than 10Gbps NAT with large packets. And i assume that the processing is then 10 times heavier but do you have any data from another CPU showing its performance?
How about some layer 7 / application firewall capability either built in or via an addon?
That'll be possible via software.
Hi Tomaz. When you predict to be available?
I tried a Qotom C3758R with 4x10GbE SFP+ and 5xI226-V and unfortunately all I226-V NICs are triggering Hardware error from APEI Generic Hardware Error Source: 0, AER: aer_status: 0x00002000, aer_mask: 0x00002000, AER: aer_layer=Transaction Layer, aer_agent=Receiver ID and the driver is crashing. Sometimes is just triggering the Hardware error and sometimes is also crashing. I have no clue if is PCIe issue in CPU or something else. I can't believe that all 5 NICs are throwing the same error. These errors are triggered on iperf3 --bidir. I couldn't manage to trigger them if I'm doing only one-way traffic.
I really need a good device and your looks very promising.
After watching this video and the previous one, I still don't understand the choice of an ARM processor. The most likely candidates for firewall softwares on that new device would be pfSense, OPNSense and VyOS and none of them support the ARM architecture yet. I don't think that hoping for those product to release an ARM version by the time you put your hardware on the market is realistic.
If you don't have a killer app for your hardware, it will not sell.
Very few people will be willing to pay more than 400 USD for a hardware just to be able to run OpenWRT on it.
What about the Intel Atom C1110?
Please Please Please make both 10G ports SFP+. You can always plug in an RJ45 SFP. But if you don’t have 2 SFP ports you can never have Fiber in and Fiber out. This is my need.
Yep. Already done! th-cam.com/video/GKj5P4goBS4/w-d-xo.html
Would I have 2x 1GB and 2x 2,5 GB
Is it possible to upgrade the 1 Gbit Ports to 2.5GBASE-T and PoE? There isn't much point in having a 10Gbit Internet or home server connection if all other connections only add up to 4 Gbit in total. And 2.5GBASE-T together with PoE is realy great for connecting a couple of Wifi-6 access points across your home or office.
2:15 with this cpu, this will be a software-router? No ASIC? A ddos attack with a lot of PPS will consume all cpus with IRQ? This remember me a Mikrotik 😅. Thanks!
LS1046 (our CPU of choice) has something called a DPAA, meaning it has dedicated hardware for a lot of network functions. So it has best of both worlds, for an acceptable price: 4 general purpose ARM cores and the specialized DPAA.
don't use 5v as input, please support proper usb-pd so that you can use any power supply, otherwise it makes 0 sense and it would be better have a barrel jack. 15w for usb-c power supplies are meant to be delivered over 9v and not 5v for most power supplies
Yep, that's already planned. So many people "voted" for it that it was a no brainer.
@@tomazzaman I just don’t want to see another raspberry pi situation where you need a specific power supply capable of 3A+ over 5v while you can buy proper usb-pd of 35w for 20€
Did you consider supporting PoE on the gigabit ports?
I wish that you could add 4 x 2.5Gbe ports instead of 1Gbe. But i guess yoiu cannot, so at least try to make the 10Gbe port, backwards compatible with 5/2,5/1 Gbit speeds. I wish you the 2024 bring your dream alive.
Thank you! I’ll do my best!
DPDK supports LS1046A. Will this hardware be validated for DPDK? Will you support upgradable RAM and storage? If I can run Linux with DPDK and upgrade to 64GB, and upgrade storage (m.2 pcie or at least mSATA) this would be great for my software project.