I would pay for this extra speed, now the how much....would have to see the full picture. If (hypothesis) the device with 1gbps ports is 170€ and the one with 2.5gbps is 200€ doesn't need too much thought to pick the 2.5gbps. But if we are talking about 50€ or even more....then maybe a model with just the 2 10gbps would seem better 😂
as i said on previous video i dont see a reason why to have a 2.5GbE if its not like a free (no losses thing, maybe if it replaced 1 GbE ports as it could still server as management but its a poor upgrade choice) since there are 2 SFP+ 10Gbps cages which can with simple modules be turned into any 1GbE and up rj-45 or SFP. also i do not understand why use 2.5GbE port anyway when most home networks from router to core switch can easily be covered by single SFP+ cage and at that speed its very likely that incoming wire will eb fiber which makes RJ-45 port kinda useless as fiber will work with SFP+ much better
@@bigpodi agree ... if you need more flexibility you can always add a switch and utilize a couple of vlans to create more ports ... if posible add a nother 10ge port instead for added bandwith for the trunk, and let people split it into 1, 2.5 or whatever themself 🙂
I think using a PCIe based solution would miss the point of using that specialized SoC‘s capabilities. Rather keep the 10GbE SFP+ Ports since most 2.5GbE Switches have 1 or 2 SFP+ ports too and you’d have to do way less on the software side when it comes to drivers and such.
My thoughts exactly. This is a router not a switch. Your internal devices don't need a dedicated 2.5GbE connection to the router. WAN -> Router 10GbE Port 1 (if you have greater than 1GbE WAN) -> 10GbE Port 2 -> LAN Switch with SFP+.
It would make more sense to have two SFP+ ports for the 10Gb ones. That way you can do DAC, Fiber, 10Gbe, 5Gbe, 2.5Gbe, 1Gbe, etc. Whatever SFP+ module you want, you use.
Whatever the verdict. I definitely like the approach of asking your audience. Hopefully, some other IT RU-vidrs can share their thoughts in the comments
@@KennyMacDermid it's a balancing act. I want to hear what people have to say, but ultimately, it's my decision to do what feels right with all the information on hand. Time will then tell whether it was the right one.
Since the router will have dual SFP+ just allow customers to buy a cheap SFP+ to 2.5GbE adapter. They are about 15-20 euros. You could even offer to put one in their shipment on your store by getting a good deal in bulk from a supplier.
I'll be super happy with two SFP+ ports already. Anything more is really a bonus tbh. I'll just be sad it's not gonna be available for me to deploy when I have to set up a network in early summer
I like the original idea of the 2 SFP+ and 2 1G ports, all using the SoC's infrastructure. A switch connected to the SFP+ ports can provide 2.5G if needed.
That's exactly what I was thinking. 10gb SFP+ to 2.5gb switches are like 60 bucks these days. Dont over complicate the router for 1x 2.5gb port for the same cost that a switch with 5-6 ports will have.
@@Allhopeforhumanity popular options on amazon: $65 = 1x10G, 8x2.5G (uses a 10G port) $45 = 2x10G, 4x2.5G (uses a 10G port, but gives you another for chaining) the total bandwidth is reduced by a whopping 1.5G by not having a 2.5G port on the router, which is not super noticeable with 10G downlink, and you can get more 2.5G ports for around the same price IF you want/need them
I think having a SoC specialised and with hardware acceleration enabled router is much more important than using PCIE to add 2.5 GbE interface. I rather have two 10GbE SFP+ interfaces, which I can add to a switch, than having a 2.5GbE interface. I would rather buy a 2.5GbE switch - as I remember mikrotik has published a latest switch with multigig capabilities.
Frankly, with a good switch and VLAN support, I think just 2x 10gig would be more than enough, especially if NIC teaming and VLANs can be used on top. If anyone wants the 2.5 port, there needs to be a justified actual use case for it beyond "bigger number go brrrr", especially if you have to sacrifice hardware acceleration to achieve it.
Well, one good and possibly common enough use case is ISP + switch + NAS, that makes 3 high speed devices. Another one is a home server, if you have a server that should be accessible form the internet in your home/office instead of renting a VPS, then you might need a third high speed port. Yes, those devices might be connected via a switch, but they both might need package filtering so connecting directly to router would be better (and faster).
There's one more option: To use key E socket and put 2.5gbe card into there. But that means there needs to be a hole for the port in the back of the router, just like Lenovo Tiny's have. I just bought 2 of those E-socket ports and I will be testing them soon in my Lenovos. I think I paid ~12€ each from AliExpress including shipping..
Hell Yeah!! CCIE here. Can 10G ports run natively at 5 and 2.5? Or are they SFP+? Got cable modem yesterday and it only supports 2.5Gbe. That means to use a 10gig SFP+ port, have to buy converter that is like $170. Or I can only get about get 950mbps of the 1400 mbps or 2000 mbps Comcast currently provides with a 1Gig. Now, a few modern cable modems come with 2 ports and LACP. Not sure if your 1Gig ports will support LAG/Etherchannel/LACP … but the 2.5gbps cable modem costs $130, and dual one gig LACP cable modem costs $300. As network engineers, we know a LAG of only supports multi-stream 1 gig, while a true 2.5Gbe supports native stream up to 2.5Gbps. Thus 2.5 just makes sense in the USA, as Comcast is the largest residential ISP, Spectrum and Cox (the other big ones), will all support over 1Gig with Ethernet handoffs, not SFP+. Thanks!
Most ppl interested are going to have 10gbe networks, with managed switches, 1gbe and 2.5gbe clients will be downstream. That said a 2.5gbe or even 5gbe port for wan would seriously future proof this device and make it very flexible indeed. My main interest is this device is that it can route between networks (vlans) at 10gbps using the sfp+ cages. Sounding interesting so far, good luck 👍
Upgrading from a 1GbE port to a 2.5GbE port doesn't seem worthwhile to me, given that I already have a connection at home that's faster than gigabit. However, transitioning to a 5Gb port would turn it into the ideal router for me. Having a 5Gb port for WAN while still retaining two 10G ports (perhaps even with the potential for link aggregation) is the router of my dreams. I'm not particularly concerned about the 1G ports, though they could prove useful for "miscellaneous" APs, such as a separate IoT network or similar setups. Equipping it with 2xSFP for LAN usage and 5G for WAN usage (I'm not sure if we've discussed potential software options yet, but most open-source or advanced router software can accommodate such configurations) would, in my opinion, make it extremely future-proof. I'm unsure about the price disparity between 5GbE and 2.5GbE, but in my humble opinion, 5GbE would make more sense, especially if it can operate in a 2.5GbE mode as well.
At those speeds I would imagine the infrastructure is already fiber (for WAN) and it makes more sense to go directly into the SFP+ cage with an optical transceiver rather than having another device to go from fiber to copper just to use the copper port.
@@vladconut You're absolutely right. It would make much more sense to go directly into an SFP+ port. Unfortunately, my ISP thinks differently. When I attempted to bypass their device using a GPON transceiver, they contacted me within 2 business days, stating that their monitoring showed something was wrong with their device and they would replace it. When I explained about the GPON module, I was redirected to someone else who informed me that this goes against the Terms of Service (ToS) and if I don't use their hardware, they will terminate my contract. So, yes, it would be much more sensible to connect directly to an SFP port, but regrettably, it's not something my ISP allows or tolerates.
For where I'd use a box like this, I would be happy with 2*10 and 2*1: This allows me to build either fault tolerance on the WAN, or backup WAN, and fault tolerance on the LAN side. A single 2.5G port is okay if all you have is a single AP, but I have many, and they are PoE, so I need to have them hanging off of (vlan tagged) switches anyway. I would be far more interested in having space for doing packet captures than onboard wifi - Where the router is, and where the APs are, for a network that warrants a 10GbE capable router, is usually different. This is from viewing this more as a business/network admin's home/enthusiast-box, and not a cheapest-isp-cpe-solution.
I saw that top comment in the previous video, very soon after the video went live. I assume it's a "give me all the ports" mentality. However 2.5GbE is an extremely new implementation on AP's. I personally wouldn't waste such a high speed port on the theoretical max speed of a WiFi AP, but if that was the use case then PoE would definitely become a must have too. Considering it's powered over USB-C PD, that could ad additional complexity. More to the point, many people (myself included) like power efficiency. And if it's going to add more CPU overhead, I would say "Keep it simple stupid" But that's just me
I definitely agree with you on pretty much everything. I just wanted to entertain the idea since so many people voiced their opinion in favor of the 2.5GbE port.
I disagree, as I just moved my whole network to 2.5gbe with fsp+ option if I ever find time to drill some holes for the new wires. All the moderately priced motherboards and mini PCs already come with an 2.5 gig port. Also my wireless routers are now using 2.5 gig link, not to mention switches I bought for ~30€ pcs. So for me the choice is either to get sfp+ modules for 2.5 gigs, or have those built into the device, or to use M.2 E-key card for it. I think the last one would be the cheapest option if there would be a hole in the back for the extra port.
@@mahagr78 There's definitely an argument to be had on both sides. There's certainly nothing wrong with adopting the newest technology. I'm glad you mentioned the M.2 E-key as I was wondering about that. I know you can break it out to PCIe on the normal M.2 but I wasn't sure if there was a direct 2.5 GbE module. I recently bought Google TPU (M.2 E-key) after watching the Hardeware Haven video about them. He showed a SATA module too but I wasn't sure about 2.5 GbE. I would just presume, if I went full 2.5 GbE, then I would be using a switch for it. Unless that traffic needed to be routed
@DanielVidz The card was a bit tight fit for the Lenovo Tiny m710q, but I was able to insert the module and fit it to the serial/VGA port slot in the back. The nic is showing up in kernel logs as "r8169 0000:02:00.0 enp2s0: Link is Up - 2.5Gbps/Full - flow control rx/tx". Also my router sees it correctly as 2.5 gig adapter. Now I just need to properly configure it in proxmox... But yes, I think the best option here is to use an optional M.2 card and just drill the holes needed for both wifi antennas and an ethernet port. In that way, people can easily decide if they need wifi, 2.5 gbe or something else. Then just provide people the options to choose between those (or none).
The more I think of it, I would not include 2.5 GbE. I would instead have that M.2 slot and make sure that both Wifi and 2.5 GbE works with it and maybe sell those modules as extra options. You can buy 2.5G switch for 30-40€ and hook it up with 10G sfp+ for another 15€. Or you can buy M.2 A+E key RTL8125B card for 12-15€. To me both of those options make far more sense than raising the price for everyone to get just one extra port, which may or may not get used. 4x 2.5G + 2x 10G port switch and ftp+ cable = +45€ or M.2 RTL8125B card = +15€ or everyone pays 40€ for a single port
Hi Tomaž / @tomazzaman, not sure if you're checking comments but the E slot does not require 2 PCIe x1 lanes. It can be wired that way but not all manufacturers use it wire the slots up that way. I actually made the change to the table to try to clarify that back in late 2022. You could potentially add an additional PCIe lane to you implementation if you wanted.
The more recent cable modems have a 2.5Gbps port. If the 10Gbps ports are not NBASE-T (i.e. they only do 10 or 1Gbps, and not 2.5 or 5 Gbps) then those cable modems will connect at 1Gbps to the router. Which is only an issue if you pay for a plan that's 1Gbps or more, like Comcast's 1.2Gbps plan. However, there are relatively inexpensive switches with a handful of 2.5Gbps ports and a 10Gbps uplink port, which is a workaround (i.e. Translates between the cable modem's 2.5Gbps and the router's 10Gbps port). Some of those switches support VLANs, so you could turn that single 10Gbps port into 4 x 2.5Gbps ports, isolated on individual VLANs. So I think it's reasonable to skip the 2.5Gbps port onboard, particularly if it would bypass the hardware-accelerated packet processing. The external 2.5/10Gbps switch is a solution for those who find themselves in that specific set of circumstances (modem with 2.5Gbps port, plan >1Gbps).
Can't you just put another M.2 E-Type Socket on the board but wired in PCIe 1x1? (Or maybe have a switch/BIOS option to disable one of the sockets/reroute the lanes to the other socket for a full PCIe 1x2?) That way people could just plug in a M.2 2.5GbE card at full speed or both it and WiFi at reduced speed after the fact, if someone wants to do that, i've seen a lot of people mention they would like to have a cut-out in the case so they could plug such a card in themselves if they chose to do so, all without impeding the 1GbE supply correct? I would love to know what you think about this approach since technically you don't have to increase the cost compared to including the 2.5GbE phy in the first place unless i missed something.
I love the name of the product!!! Something that maybe a lot of people forget is this is a router, not a switch; keep cost down with the two SFP+ ports and *maybe* another copper port for easy management. Also please consider adding a M.2 for NVMe storage even if only connected with 1x PCIe lane.
I’d want two M.2 for mirrored boot drives. Wouldn’t matter whether they were SATA or NVMe But unless you’re talking about DNS, DHCP, or other similar services that run within the routing/firewall software, I’d rather have any other services running elsewhere. I like to let my router route, my NAS serve files, and everything else run on a virtualization host of some sort. YMMV Edit: A console port would be very nice to have.
@@vladconutI wouldn't recommend running docker directly on your router. Docker has a ton of security issues and from my experience running it on anything that is not behind a firewall is just asking for trouble. We had servers hacked many times due to docker messing with linux firewall and disabling security rules. There is even a well known bug that docker (the company) does not want to fix due to "compatibility", where if you set docker to run only on one network interface it will sometimes ignore that rule depending on how you started your container. TLDR: Don't run docker on your router.
M.2 slot doesn't have to implement all pcie lanes, just one is enough unless you want more speed. And since you aren't likely to require more than 8 Gbps of bandwidth for your router - it's not a NAS - you can get away with implementing just a single pcie lane in the port. Infact a lot of embedded systems do just that.
I wanted to say that, Also you don't need to implement all the other interfaces. There is one gotcha, though: some devices like eg. intels network/bluetooth combo cards use both lanes since they are just 2 separate devices on a single board.
Remember that your firmware can allow multiple Serdes configs on the same hardware. For example, you could allow a software-downgrade to mode 2233 on serdes 1 to allow people to plug in 2.5 Gbit SFPs into the 10Gbit/s SFP+ Ports if they have no 10Gbit infrastructure - usually, this is pin-compatible or requires minimal change to the pcb for dual use compatibility
(although probably less sensible overall,) the same could be done for SATA - move the M.2 to RX2/3 on the second serdes, and also route rx3 to a SATA Port, now the user can choose between using the M.2 at x2 or the SATA port and M.2 at only x1 - although here it will require a bit of extra hardware and good communication so users don't think they'll be able to use both, so that might not be worth it
I commented previously about a 2.5gbe port but I’ll be perfectly fine with the proposed layout given the cost increase. Thanks for the update! Your wife’s advice is spot on haha, cheers Tomaž🙏
Why do you even need 2.5gbe when you already have 2x10gb ports? Would it not be better for you to plug a switch into one of the 10gb ports and divvy that up as far as you'd need?
I'm voting for keeping the 2x 10g and 3x 1g original config. Use a SFP+ DAC cable and connect to a switch with 2.5g ports. Much more flexible and use VLANs to segregate the traffic there . Maybe sell a bundle with a switch like this and an AP?
I seem to be in the minority but my use case a 2.5 gbe Port is useful as a WAN. Around here if you want multigig internet with Comcast your options are either LACP or 2.5gbe. If the sfp+ port can negotiate down to 2.5 gbe that may be an option but I personally found many sfp+ ports which won't run 2.5 gbe even with an adapter and instead not working at all or dropping to 1 gbe.
I was skeptical when you said you weren’t even planning to support wifi, because of the dozens of devices I have, my Raspberry Pi is the only one with an ethernet port (making the 2.5g vs 1g question irrelevant to me). I’m really pleased with your slotted approach, though… this could be the best-value router (ever) if it has the ability to be upgraded over time with new wireless modules. It sounds like you’re doing a great job of evaluating the trade-offs and finding ways of being price-competitive. I hope you can sell the router both with and without the wireless card so that people can opt in if they need it or can buy “verified compatible”modules and upgrades from you in the future
Pozdrav iz susjedne Hrvatske. Svaka čast na projektu, moje mišljenje je da je 2x10G i 2x1G i više nego dovoljno. Ako nekome baš i treba 2.5G uvijek se može staviti neki switch i povezati ga optikom ili DAC kabelom na ruter. U svakom slučaju ću kupiti ruter jednom kada je dostupan. Da napokon nešto nije made in China.
Wouldnt you be able to sell a seperate version for those interested that comes pre-shipped with a pcie 2.5g port adapter and respective mounting point / cutout in the case? This way you don't lose hardware acceleration but can still offer a 2.5g port for those that want it. You could even have just one / the same case for everyone, but just have a cover for the unused 2.5g port. This way if somebody wanted to upgrade down the line, they could add it themselves (And you could sell the module for it seperately)
I have super simple homelab with opnsense running in a proxmox VM, but I’m se fascinated by the design process and all the videos you are putting out that I’m definitely buying one of these as soon as they become available. Such a cool project. I just hope you’ll ship to Canada so I can buy one! As for my opinion! I don’t think it’s worth it. The 10 gbps ports are fine!!
You could also let the 2.5gbit be an optional extra over the m2 slot. Yes, the user had to decide between WiFi or 2.5gbit Ethernet. With an “optional” module that you can insert in the m2 slot and a provision for the optional Ethernet port in the case you might have the most flexible solution. The cost of making the provision for the Ethernet jack in the housing should be less than the cost of including the 2.5gbit port in the base board. This would give you 2x10gbit, 4x 1gbit and optional 2.5/5gbit Ethernet or m.2 WiFi.
I already have a different solution in this performance class.. However based on the previous video picking the CPU based on a 7 year support window without change, I can only say that 1Gbe interfaces will look ancient once WiFi 7 has been the most common WiFi type for years (Thinking 2 years into the future from now). I think most users however will just be able to use the SFP+ ports with a switch to deal with this. Alternative configuration idea.. Make the SFP+ ports SPF+ OR 10GBe.. IE you have 4 physical interfaces but can only use one or the other for each "port". 10GBe SFP+ modules tend to be very hot so for those that want 2.5/5/10Gbe use but not SFP+ this could be a good compromise. You can see this done on a number of 10GBe switches these days like the QNAP qsw-m2108r-2c. This might be a cheaper way to have greater than 1GBe ethernet ports on the device.
I would vote, but I'm not currently in the market for a router. I watch your videos to gain some small understanding of what it takes to bring a product to market. In this, you're doing a great job!
I would appreciate a 2.5Gb port for WAN connectivity and would be prepared to pay for this feature. While it's still very uncommon for ISPs to offer gigabit speeds in Europe, it's not the case across the Atlantic, and Unifi have understood this very well. So, to sum up, what I'm looking for: - 1 x 2.5Gb: WAN - 1x 1Gb: another LAN (or a wan failover, with reduced bandwidth, but this is for temporary failover) - 1 x 1Gb: Management LAN - 2 x 10Gb SFP+: uplink to layer 2 switches (much cheaper) configured in LACP (Link Aggregation). These can be POE+(+) so no need for POE out on your router. - On the software side, compatibility with PFSense, please. In fact, you may succeed where Netgate failed with its 4200 model (which lacks both SFP+). Also, do you intend to publish the performance that can be expected in routing thanks to this CPU (as Netgate does)? Thank you and good luck with your project. I'm following it closely...
Same for me dual sfp+ the 2.5 I get it people want for AP’s but that’s pointless without PoE. I’d rather some extra go towards supporting PoE in or alt power. Also FYI the OSI layer model you use is the obsolete 7 layer one that never became anything. TCP/IP and packet switching won that war and OSI was reorganized to the 5 layer model that represents networking as we know it so they could salvage some of the effort they had put in.
Awesome video, awesome presentation and research as always! Honestly i don't see the issue with not including 2.5gig. This is a router, not a switch. if someone REALLY needs 2.5 gig, (which isn't a lot of people, because anyone who wants high speed NAS access or high speed between PCs in general will use 10gig and everything else like WAPs or cameras or whatever all use 1gig.) there are plenty of switches with one sfp+ 10gig port for communication with the router and a bunch of 2.5 gig ports for all the devices. Just adding needless expenses and complexity to an already premium product seems rather pointless to me for such a niche usecase.
Having a 2.5Gbit port is always nice, but this is quite the drawback. And replacing all 1Gbit with 2.5 would make the whole thing cost a whole lot more. That makes me wonder now though... if it is an issue of mix and match, would there be a huge problem with compatibility if you just axed the three 1 gig ports and replaced them with three 2.5 gig ports? I'd wager that's not the case. Quite the conundrum. That being said... I think the 10Gbit can easily handle anything with 2.5 gig capability, and you can use switches to have more devices (albeit they'll need to be better switches). So maybe not having a 2.5 gig port natively is not that big of a deal.
A faster third port could be great for implementing a DMZ for home usage or a guest network. Self-Hosting should be a typical use case especially in the target audience for such a router. A layer-2-switch could split the 10 GbE ports as required to supply more connection options, but configuration of a DMZ would require looking for specific VLAN capabilities again, driving up the cost immensely. I think having a more powerful third port would drastically increase the use cases for the router, but I‘d be interested in hearing other takes on this, if someone would like to comment. It‘s unfortunate that this would undermine the specialized SoC by necessitating PCIe though.
@@Luniii737 I can understand that, but apart from the increased costs, the drawbacks only relate to the upgraded port, so unless the three 1 GbE ports are all budgeted in one‘s setup, the drawbacks would only affect those who wish to use the port. I don’t like to rely on VLANs for segregation in home setups - for one, they need more expensive hardware, and they are less secure than physical ports, even though most known exploits can be mitigated. I think having another capable port for a different network directly on the router would really future-proof the device.
I would say stick with the initial design I don't know about most other people but as long as we can configure any of the ports and WAN or LAN. Then I don't see a reason to have the 2.5G NIC. I know here in the USA most people have a 1G or less WAN connection to the internet and for the ones that do can just use the SFP+ port with a adapter transferer to go from fiber to copper. That would also still free up the second SFP+ to use for downstream LAN connection to a distribution switch or the core switch in the home. So I would say just stay with your original design and if enough people buy that one and still want more connectivity just make a version 2 with a focus on higher speed ports and no 1G ports.
my inclination would be to keep the ethernet configuration as-is and use the left over lanes for a M-key M.2. Putting a low cost PCI-E NVMe (like optane) will make the router much much more reliable and allow swappable storage.
Multiwan for the simplest thing I can think of on the top of my head! Other one could be to connect a guest's AP completely isolated from the rest of the network. Would be also good if it is POE but I can live without POE.
In my mind it's likely that the people who are ravenous about it are in the homelab/"smb" space and most likely not willing to pay a fair price for the unit. If I'm right they are thinking things like "2.5 is more than 1!" "I have a new 2.5G desktop/server and don't want to lose performance" "It's expensive to upgrade to SFP. I won't do it until it's cheap". To give them the benefit of the doubt, it's possible some have internet connections greater than 1Gbps and they're fine with their existing network equipment but they just don't want to justify investing with that initial cap due to Ethernet overhead. In that case SFP in and 2.5G to the LAN is a fair compromise. But even then what they should be doing is upgrading to a switch with SFP in
This feels like a classic "if I had asked customers what they wanted I would have made a faster horse" situation. People who aren't putting money down can lead a product in bad directions and even some of the people who do put money down are not right.
If it really came down to it, I think I'd be alright adding an SFP+ module that can negotiate at the various NBASE-T speeds to the router's included SFP+ port. I don't think the price increase is worth the very slight reduction in functionality, especially if that locks out access to the PCD part of the silicon. You could also (potentially? I'm not speaking from experience here but from what I've seen in the videos already) very slightly increase power efficiency if packets can all be processed by the same core, leaving the other cores open for other tasks. Maybe it would be worth experimenting with this. I have absolutely no personal nuanced experience and I am by no means an expert, but gosh darnit if I paid for the whole silicon and all those features? I'm gonna USE the whole silicon and all those features!
Hi Tomaz. Question and idea: AQC115 is 2.5GbE single port. Why you should waste these PCe 3.0 2x lanes just for one 2.5GbE port? Yes, is increasing the price of the router a little bit, but adding a PCIe 3.0 switch on 2x lanes or even PCIe 2.0 x2 lanes and distribute to 2x or even 3x 2.5GbE PCIe NICs can be a solution: PCIe 2.0 switch 2x lanes has a bandwidth of 1000MB/s (8Gbps) and 3x2.5GbE should have about 937.50MB/s (7.5 Gbps). Also if you use a PCIe 3.0 switch you can use some additional lanes for other devices. Or in my opinion you should optimize the full number of ports like this: 2x10GbE and 3x1GbE interfaces using SerDes1 0x1133 and SerDes2 0x5A59 or 0x5A06. In this case you will have 3x1GbE interfaces + 2x 10GbE via DPAA and additionally you can add a PCIe switch Gen3 to add 2x 2.5GbE interfaces which obviously will not go through DPAA. But PCIe 3.0 1x can handle 7.880 Gbps bandwidth which is more than enough to use for example 2x 2.5GbE interfaces via only one PCIe 3.0 lane shared via a PCIe switch. I don't understand why to waste PCIe 3.0 x2 on a single 2.5GbE chip. In SerDes2 0x5A59 also enables you one SATA if you need.
For me, I only run 10G fiber or 1G copper at home and nothing in between so it wouldn't be of use for me, but I feel it would be a very popular addition with many people. Love seeing these videos it's so interesting to see the technical aspects of the project
2.5G is a must imho,I even Vote for 5Gb addition. New gen of MBs for am5 and lga18xx probably gonna feature revised realtek 5gbs chips. Most of the mb buyers probably wont need that kind of a premium device but still it will be there while others may not. All the best!
Absolutely, add a 5 gig. Or, have it and the 2.5 gig as optional add ons for a “made to order” approach, I will definitely be buying the 5 gig option though. Out of curiosity (or cause I just forgot) are you adding the ability for a cable hookup or will it just need a media converter for the people stuck with one ISP?
That differential pairs idea is from the telecom world. I use to lay cables that did just that with the signal and it was explained to me exactly the same way you did 🙂
Awesome video and enjoying this series a bunch! I have a tiny grammar correction for you, but I’m mostly posting it to sympathise with how hard English must be to learn. Around 9:50, you say “the good news don’t end there”. Despite etymologically coming from the plural of “new”, “news” is singular, so you have to say “the good news doesn’t end there”. English is a silly language sometimes. Struggled slightly to word this, it’s more me being a linguistics nerd than me trying to correct your grammar, the meaning was obvious from the words you said.
In my target setup the router/firewall will be connected to an managed switch with one or both (LACP) 10G (SFP+) ports while using VLAN's I need 1 or 2 additional [Slow] NIC's, 1 for "High availability Sync" to the second router, and 1 for management (optional). (even 100M will be enough) So personally i don't care about a "faster then 1G" port.
I have an alternate solution, it looks like acquantia xfi phy ics, you could put a serdes mux between this and one of the xfi serdes, and then use this to implelent a 2 5g ethernet combo port for about $30. Though probably the better option is to suggest people use a switch or a 10g-baseT tranciever that supports 2.5G.
I initially supported the idea of a 2.5GbE port, but seeing the implications of adding it, I realize that the original idea of two 10G ports and some 1G ports at a more affordable cost is better since you can easily get 2.5G with a switch. The implication of not having dedicated packet processing hardware available for the 2.5G port is important too. I would go as far to say I'm not even sure Wi-Fi is the job of the router. After all it is a router, not a router + access point combo we normally see in low end off the shelf "wifi routers". If I was spending this much on a quality router, I think I will spend on quality Wi-Fi access points too.
My thoughts exactly. I'd rather make an awesome router that's good, well, at routing. And delegate the rest to other devices. Such as switches and access points.
RTL8372N are like $10 and have 2xXFI and 4x2.5G ports, if the SoC supports the Realtek DSA modes you get all the ports you want and proper interfaces on the os side.
Also, separate idea, modular IO? I don't think you could do the PHYs as a socked thing, but, maybe a IO module for different setups? You can still make/sell only 1 to take advantage of scale and discounts, but you could open-source the module design and approved parts/BoMs and let the crazy people take the PCB schematics to PCBway or something and let them take that cost on themselves?
Well I mean you could get sata to pcie host ic But yeah SATA sux, also added cost. Someone is determined they can use M.2 M key or even E key to add sata (one of those exotic adapters from china).
The best bet is make the 1 Gig Port standard and in the board itself make the chip that supporting port swappable like a socketed CPU. That way the order can be full filled with more flexibility
WIth this explaination you did I guess 2 x SFP+ will do, and we'll see if we need the 1 gbps. I will just connect the router to my switch. just try to make it so it does not overheat when using 2 x 10 gbe SFP modules if someone can't use fiber modules or DAC cables. After all you can make 10 Gbe ethernet ports using SFP+ but not the other way around.
Well, 2 10gb ports connected to a switch with more 10gb ports does the trick for the ones that need more then 1gb port. It's more expensive then you including a 2.5, but way more versatile. By playing around with profissional FW/router's every day thats the most common setup, fw/router link to a switch with 20gb/50gb trunks and then the switch does the rest.
A Ubiquiti UDMPro runs $379.00 with 2-10GB and 1-2.5GB so if your device was up to $500.00 with those specs I would totally buy one. In fact i would gladly put a down payment on one or buy it on kickstarter.
The real shame is that 2.5G is only used by CPE manufacturers that create retail consumer devices. 10G is far more abundant and available not to mention less expensive than 2.5G. In most cases people only need a single 2.5G port to accept a WAN handoff from their provider, in this case just use a 2.5G to 10G SFP+ media converter and keep your beautiful router filled with 10G/1G ports as the universe intended.
5 gbit/s or 2.5 Gbit/s yes. I already have 4 Gbit/s fiber at home (so... maybe a good WAN port?). EDIT: SFP is also fine instead of 10Gbit/s UTP. EDIT EDIT: Nevermind, maybe I don't need those 1 gbit, 5 or 2.5 gbit/s ports. It could be that my media converted can be connected to SFP+ port via a SFP Transceiver Module. And then I will add 10Gb/s switch behind the router. Does that make sense?
because you're not going to have a whole bunch of fast, poe-enabled ports, the audience for this is going to have a giant switch. its fun to imagine having a 2.5G wan with each 10G port feeding its own switch, but there's barely even gigabit service in my city, and that wouldn't be worth the constant extra cpu overhead. A 4×2.5G POE + 2×10G switch is like $100 nowadays, so it's worth it to make the router more focused. Sticking to features enabled by your CPU will make development easier and faster. By the time version 2 is in development, the cpu and phy chips will be cheaper and faster.
To make it easy: If it's to much work i.e to many downsides, then No. If it's possible which out too much of a hassle, then yes. Also: I'd take the price increase anyday for the 2.5Gig but only if the other feature sets are stil possible
Its a router, not a switch. Makes no sense to add a 2.5. If it was 4 dollars, and used the HW-accel, sure.. but its not and that just makes it all idiotic. Feature-creep until will kill it.. someone will soon ask for a HDMI out and a pci gen 4 for a 4090 for Plex transcode
Ditch the 2.5Gbe and add another M.2 slot as it is more flexible. Users can add their own M.2 to 2.5Gbe adaptor if that is what they need. Maybe allow a blank punchout in the back of the case where the RJ45 port from the M.2 to 2.5Gbe adaptor goes. For examples of these adaptors watch the video from Wendell @Level1Techs
Yeah do it, 2.5GB is the way to go, here in Portugal there are already 10gpbs connection for 100euro a month for private individuals. So 2.5Gbps is the least, 1gbps is to old, especially if your using this for Vlans or other
I'd stick to the original design as many commenters have said but think about allowing users to put a 2.5G interface card in the m.2 slot. Should only be a consideration of having a blanked off cutout to install the port (see KIKAPA M.2 A+E KEY 2.5G Ethernet LAN Card RTL8125B). Whether or not that's a good idea or not (realtek chip in that example) I'll leave to those who have a better idea.
Note: I wont be purchasing it, but as someone in the ISP/Network field - here is my take: Don't put the 2.5Gbps port in. Check and confirm the 10G can negotiate at 2.5/5Gbps (should be able to ... but never assume) Why? By using the PCI lanes, you now push all traffic over the CPU, increasing latency and if people are going to be using this router to do more than routing (most likely) - then you can incur random bottlenecks in the network throughput. Specifically at higher speeds - CPU based network processing is ineficient compared to a dedicated chip. If ppl want the ethernet port - then have a look at combo-ports instead. Basically wired the same way, but you get an sfp cage and a RJ45 - uses more space, but lowers cost and increases ease of use. Since if someone wants > 1G but has ethernet ... just use cable. But if they want to use fibre, or a DAC ... they can - it becomes a choice. That way you then would have 2 gbit (or 4?) and 2x 10G avail ports (either sfp or ethernet) In home lab setups - a 10G switch would be used .... so no issues in small homes - probs 1G for internet, leaving 2x 10G internal for the middle section that kinda wants faster speed, but not sfp style ... they can just use the ethernet jacks. 2 ports is fine and covers 80% of the problem. And if the 10G can drop to 25/5 ... then irel overall. And if someone needs more ports .... get a bloody switch - a router is a router, not a switch -_- haha
1 10G for incoming, 1 10G to a NAS and the 2,5 or 5G to "master/monster" editing machine, rest for subpar devices :D I'd love something like this. Or 1G for incoming/WAN, 10G ports to NAS and powerful PC, 5G to Waifu/gaming/secondary PC or a switch with all the IOT and other trash.
@tomazzaman please make sure that the linux kernel can handle forwarding 2.5Gb of traffic. from my experience, there is an issue with linux on arm when it comes to high data rates, even 1Gb puts a lot of load on the cpu.
Not this CPU, it has dedicated networking hardware. Even the two 10GbE ports should be easy. I'll report once I test the development board (in a month or so)
I think two 10 gig ports ought to be enough high speed ports. if someone really needs 2.5G due to wiring constraints, they can run one of the 10 gig ports through a switch.
personally, i prefer 2 SFP+ at low cost, switches with 10Gb/2.5Gb are numerous and cheap on chinese sites. moreover if it's SFP+, there are SFP+ modules capable of 2.5 or even 5Gb. In conclusion, as cheap as possible for the brain, distribution switches can easily compensate for this need.
Your mention of losing DPAA hardware processing by moving 2.5GB over to PCIe is what kills it for me. I'd rather have integrated support for 2x 10GB SFP+ and 2/3x 1GBe and have better hw acceleration over tacking on 2.5GBe. Like others are saying in the comments: This is a router, not a switch :) Not to mention that 2x10GB and 2x1GB is a near perfect spec for home/SMB use and is perfect for budding multi-gig ISPs. 1x 10GB for WAN, 1x 10GB for LAN, 1x 1GBe for Smart home/IoT, and 1x 1GBe for OOBE. If anything, I'd personally rather see just 2x 10GB, 2x1GB, and PCIe lanes for an add-in card down the road. I was going to suggest IPMI using something like an Aspeed chip, but for a router that really makes no sense, and it's more logical to just have a serial controller for console access imo. Love the project, will be following closely :)
You clearly didn't even described specifically who you want to make happy with this router in the first place, looking at comments some people expects universal solution, like typical soho router(so wifi and other stuff), others wants just routing and would be happy with 2x10g which routed with "asic magic", so you definitely need to clarify at least that, only then choose what specs are expected from router and only then decide what SoC actually you should get
From my experience with home networking a high performance home router ideally for majority of users needs only 2 ports but for convenience and flexibility 3 or 4 ports is probably best. I would expect that people who need a high performance router like this have a high-speed network with an advance setups with VPN, QoS, VLANs, smart firewall bouncers etc. And those users already have or will highly benefit from having dedicated switch(es) and access point(s), and they can afford those devices or find some good used/refurbished options. Then why only 2 or 3-4 ports? Because if you have a vlan-capable switch then you really need only a WAN and LAN port. But of course having only two SFP or two ETH ports would be too restrictive so it's probably best to mix SFP and ETH and that has the added benefit of allowing the users to easily create multi-WAN or multi-LAN setup and/or some service port. So to me, a router like this needs fewer high-performance (ie. utilizing dedicated SoC infrastructure) ports rather then multiple ports of many types .
What type of interface will the 10Gbps ports be? SFP+ or RJ45? In both cases you can just connect it to a 2.5Gbps switch right? I don't see the need for a dedicated 2.5Gbps port, just skip it.
I'd get TWO 10 Gbit ports if possible. Then I don't have to upgrade for a while and just need to get an additional switch or something or just hook it up to a single accesspoint.
The cost difference doesn't make sense for that little gain, leave 2.5 GbE out :) Besides, if you use PCI-E instead your traffic will NOT be processed by the Parse, Classify, Distribute logic, and will consume CPU cycles. High CPU usage means more power consumption, less efficiency, more heat, it MUST be avoided. And speaking of CPU usage, I want it to be able to allocate most of those resources for Firewall, VPN or other services; not the traffic switching. My 2 cents for the project :) Keep up the good work
