$1m server vs $600 router

We don't need all the high-speed ports. We just need the 50 gig ports that we're going to be running at SFP plus 10 GB speeds. >> Do you think a uh a Raspberry Pi would replace this box at any anytime soon? This is the Mono Gateway, a customuilt 10 GB router. I was sent this early preview version by Tomas, who's also on YouTube and who's been building it in the open for like wow 2 years now. Initially, I was interested just because there aren't many compact 10 Gbit routers that fit in one U in a min. But then I got more interested because of how the team at Mono's been sharing, sometimes in painful and excruciating detail, every step of their hardware development journey. I'll let you go check out Tomasa's YouTube channel. If you haven't seen it and you're interested in hardware design, you're in for a treat. But today, my goal is to take this thing with me to Phoenix. And there's somebody there who I know that has a really cool way to test this thing. and we're going to see if it actually gets 10 GB of throughput or not. Patrick, >> hey Jeff, >> I have this uh little switch/ router and I hear that you have ways of testing these things. >> Oh yeah, we do. >> Have you ever tested something this fast? 10 Gbits. >> Uh yeah, we definitely have. >> I I see that there's something weird here that I've never seen before. >> Well, I don't know what you're talking about, but we do have two things that are pretty interesting here. So, first off, we have a 10 gig 4 port uh switch, which is the Microix CRS 3044XGIN. I don't know how I remember that, but somehow I do. Uh, and that is a pretty low-end switch. It's sub $200. And then underneath that, we have the Nvidia SN5610, which is a 51.2 TB switch with 64 ports of 800 GB Ethernet. >> That's uh that's a little more than this. Yeah. >> And I've heard that you have a way of testing even that much bandwidth, but I don't have that much today. >> We do. Now, to do 51.2T, of course, takes a little bit of extra effort uh as you're configuring this. But what we do have is we have this over here. This is our Ksite Cyperf box. And if you don't know what Ksite Cyperf is, Ksite is a company that makes a lot of the test measurement gear that is used throughout not just the networking industry, but other industries as well. And Cyperf is a tool that allows you to do stateful traffic and do all kinds of different application traffic. So it's a way to test real application workloads on network devices just like that. So so this is just to give a little context. This is the mono gateway. So it was it was designed all customdesigned around a MediaTek chip at at its heart and it's supposed to support hardware offload. And what I want to see is whether that can actually achieve like 10 gigabits of routing. Oh yeah. So, can you show me like can we can we run this thing really quick and test it >> 100%. So, we have found devices, by the way, that don't do their wire speed on their high-speed ports, which is always kind of fun. Uh, I think this is going to be okay, though, cuz I I've heard of this project, and this looks like something that's pretty cool. What I think we're going to do though is we're going to run it on this box, but we're not going to run it on the set because just to give you an idea to run this, this is uh dual 128 cores, one 1 and a half terabytes of memory. Plus, you're going to see that we have over uh 2 terabs per second cuz we can actually generate over two terabs per second of network traffic. So, we have um you know, plenty of generation capability, but we're only going to be really trying to do 20 Gbits cuz two 10 gig ports. My thought is why don't I show you how we're going to connect this and then we're going to go put this in the other room and we can actually go run the testing and we go look at how this performs. Okay, so this is the back of the super micro server that we're using for our test suite. And what you'll see is a whole bunch of different types of network connections. And the reason for that is I just wanted an easy way for our team to go and connect the various devices whether those are switches or network gateways and connect them to our giant box, right? because otherwise you have to go through other switches like we have other 400 gig 100 gig all kinds of switches to connect but for this like you don't want to go through like three switches to get down to 10 gig. So what we have is a variety of ports including these SFP56 ports. Now SFP 56 those are the 50 Gbit per second ports but you can also run you know 25 gig Ethernet you can run 10 gig Ethernet no problem on them and so that's what we have here. These are four of the 50 gig ports and then we have another four that are right here. >> I see bigger ports too. >> Good call. Okay, so this is super exciting. So we have two 400 gig OSFP ports. So if you see these giant things here, they take these huge optics and these are OSFP 400 gig. So 400 gigs per port. Now that's important because we have a PCIe gen 5 server. So PCIe gen 5x6 is 400 gigs. So that is the idea of you know literally just saturating a BY6 link like you would have for a GPU or something like that but just for a network port. And then you'll see that we have two other cards in here which are brand new and this might be the first time we're showing them off which you'll see one here and one here. And these are Nvidia Connect 8 cards but they're dual 400 gig. So a total of 800 GBs coming out of each card. Now, you might be asking, hey, if you only have PCIe Gen 5 by6 and it only does 400 gigs, how can you actually do 800 gigs out of a single card? We had special cables made to be able to go all the way from the cards here all [snorts] the way to the other side of the chassis so we can steal other PCIe lanes. So, these cards actually have 32 PCIe Gen 5 lanes going into them so we can drive those QSFP 112 ports at full speed. I know that's like super dorky, guys, but at the end of the day, it's a lot of work. It took us months to get this box tuned and up and running and all that kind of stuff. And we're still upgrading it and making it even cooler. But for what we're doing, we don't need all the high-speed ports. We just need the 50 gig ports that we're going to be running at SFP plus 10 GB speeds. Do you think a uh a Raspberry Pi would replace this box at any anytime soon? >> [panting] >> I mean, it has one lane of PCI Gen 3. >> I know you love the Raspberry Pi, and I love the Raspberry Pi, too. We have lots of them sitting around here. Uh, in fact, we have one sitting on set right there. But, uh, no. Okay. So, what we're going to do is we're going to plug Always the wrong way. First, we're going to plug uh one of the SP cables in to our WAN. exact opposite every time. Isn't that amazing? >> It's USB and SFP. >> Wild. Wild. And by the way, we are using 25 gig SFP 28 cables here just so that way we don't uh don't have a cable issue. But then what we'll do is we're going to plug that into the second port, which is going to go right here. Boom. And then what we're going to do is we're going to go take this whole setup and bring it over to the other room where it's not going to be as loud on set. And then we're going to go and start running this stuff. and we'll show you on the screen how this looks >> a few moments later. >> So, we ran our test and it was extremely loud. So, it's a good thing we didn't have that sitting here, I guess. But, uh there were it it generated all these graphs. Can you explain what we're seeing here? >> Sure. Okay. So, Keysite Cyperf is a way to go and test a number of things and we're only using a couple of the features in this. Some of the features that we're using is we have a test that we developed after we got the tool and it's taken by the way months to go and profile these things. So, it's not like you just get the tool and you just, you know, everything works out of the box. You have to develop your own uh or they they have some pre-anned ones, too, but we developed our own set of tests. So, that way we can run all of these devices through it. And we're using two of our tests that we use on the ST main site. So, if you want to see other devices tested similarly to this, you can go to the ST main site, things that started in about Q4 of 2025, they they will use these exact tests. And what you'll see here is one example. So this is we call it the security test setup because this actually has the new security profile which we haven't shown on the ST main site yet but we're just using the straight throughput. And what we did was we set up a test where we were testing birectional HTTP traffic. So this is not just simple you know packets that we're just spewing across. We're actually doing HTTP you know post gets um and we're doing that on uh through the device. And so something you'll see here is our throughput. We didn't necessarily get all the way up to a 20 gigbits, right? We got about uh 17 and a half gigabits for our uh you know layer four and seven traffic or two and three which you know is the lower layer traffic. Uh that's at about 18.4 to 18 1.5 Gbits per second. So overall pretty good. Now, I do want to point out really quickly that when we are running this, we're not necessarily running this with like huge jumbo frames because we are just doing standard HTTP traffic. And so, what that means is that there probably is a way to optimize beyond what we have, but we're doing this as a standard so that way we can have the same test for everybody and we have the same test parameters for everybody. >> It's pretty real world, too, because most of my stuff isn't using demo frames anyway. And >> Exactly. Yeah. Exactly. And web traffic by the way uh like HTTPS and HTTP requests are pretty uh pretty prevalent in the industry these days. So uh other things that you can see here though is that it also simulates a number of users. So, you're going to see that we peaked at about uh 512 users with concurrent connections. And then we also uh you know for our steady state once we got all the way up to that uh you know 17 18 gigabit range we got to 128 gigs and it's actually testing and seeing what the device under test is doing. So you'll see that you the latency which is by the way very good compared to other devices started out and it was well under you know say 1 millisecond or so but then all of a sudden our TCP latency as we ramped up to those 512 users you can see that we also have a ramp up in terms of our latency and then we went to that steady state we were at 128 users and our latency went down quite a bit and we also have statistics on this where we can see like the packets per second we can see >> I think we got like 800 something thousand on this test. I >> don't remember. >> So, not quite a million. >> Yeah, not quite not quite a million. I think this one, but close. Yeah. >> Um, and so that's one test that we did. And then another uh test that we ran was this one over here. So, this test is a lot more complicated. And so, if you don't know this about Cyperf, Cyperf, again, is not just like Iperf 3 traffic, which is just like spewing uniform packets at a device or through a pipe. This is actually doing real applications and doing different types of user actions which is one of the more exciting things for this. So what we have here is a 11 application mix that we used to test gateways and the idea was to come up with something that you know when you do like you don't really want to do like 500 applications at once because it just becomes just a bear to do especially on small devices but we came up with an 11 app mix. So we have things like chat GPT. We're do doing Google Drive, Google Sheets. We have some just random web traffic, LinkedIn, Netflix, Office 365, Outlook and Calendar, Reddit, X.com, YouTube. And we have a number of these, you know, these 11 apps. And the idea is that we have a mix of these apps and that's what's generating different types of packet sizes, different user interactions and all that. So what you'll see here is that our throughput was a bit lower. Got to just over 16 Gbits per second here. And then also if you look down here, we got to just over 17 gigabits per second. So it's about a gigabit per second lower than when we were just doing uh just straight HTTP requests. You also see that depending on our simulated users and the connections per second, we also saw different latency. One of the cool things that you can also do is you can drill down and see. So for example, if a device was having issues with signing in or signing out of a service, you can actually go down and see those metrics if that was a problem. And why that's a thing here is because this also can inject security attacks. So as you turn on things like firewalls that gives you the ability to go and see how much of like you know malicious uh browser attacks or whatever are actually stopped by devices. So, something that we're going to start doing in early 2026 is showing folks what happens when we start turning on both the base firewall and then what happens when we start turning on, you know, premium firewall features and how many of a set of attacks actually are getting stopped by those. So, so it sounds like there's like three layers. There's IPF is what I usually do, but it's good for saying like this says it can do 10 gigabits. Can it do 10 gigabits? In the ideal perfect scenario, yeah, most things should be able to, but this is doing like actual web traffic. And then you're saying you're also going to add on security, which is another layer of if there's an active attack, can the device stop it and not kill itself while it's doing that? >> Exactly. And some of the things that we already are seeing is, for example, when you turn on those security features, you might see the latency jump up as the firewall is actually looking at, you know, what what are these packet signatures and like what's going on in this traffic, right? So, that's something that I think is really interesting and really nobody in the industry does. So, this is not a cheap setup, by the way. Um, as you can imagine, 1 and a2 tab of DDR5 these days is not cheap. 228 gig uh processor is not cheap. Plus, all of the nicks, all of the optics, and all the all the bits that we need to connect this. And then just beyond this box, we have other switches so we can break out and get to different types of ports. um because you have some NRZ, PAM 4, different different speeds on wires and all kinds of stuff. So, just to be able to do that and manage that, we have other switches that we're using. And so, this is not a cheap or easy quick thing to set up, but we've been doing it for the last couple months. And we're actually getting some pretty cool results. So, we got tons of great test data off this thing. And uh you know, initially when I saw this thing a couple years ago, I was like, I don't know if they're ever going to finish it. It's so cool to see it here and it's performing, but you're the one that knows. I guess I'm I'm used to like 1 Gbit or 5 Gbits. uh is this is this a good 10 gigabit router just based on the preliminary performance that you're seeing? >> Sure. So, I think that number one uh this certainly performs better than some of the other devices that we see. I love the fact that it's OpenWRT and you can also see the impacts of the hardware offload when we were looking at the load while we were running these tests. The CPU utilization was a lot lower. We've actually tested devices where that's an issue. So, uh I think maybe the best way to answer that question is I have already contacted him and tried to order one of these. >> Yes. And I I actually So, this is a development unit that he sent me. This is one that doesn't even have the little features that they added to the final development production run, but I also ordered one because I want to put one of these at my house and and put it through its paces there for a while. >> Yeah. So, >> this is just super cool. That's really the reason that we ordered one. >> Yes. Well, thank you so much for all your testing and uh thank you for welcoming me to your place that's a lot hotter than St. Louis right now. >> All right, Jeeoff. >> All right, so I'm back in St. Louis. Thank you so much to Patrick again for letting me uh use your networking test equipment to test this little guy. Before I close out the video, I did want to mention a few specs in case this is the only video you watch on this thing. It has a four core CPU from NXP in it. I'm not sure if that's the final final one for like the production run, but that's what's in these developer units. Uh it has 8 GB of ECC memory inside of it. So, um you know, this is it's a substantial little Linux box. It has two 10 gig SFP plus ports on the back. Those were the ones that we were testing. This one is routed to the WAN and this one is for your LAN. And then there's also three uh one gig ports. So, you know, these are kind of just a little bonus in case you need another device that's not high-speed or anything like that. Then there's three USBC ports. I know that this one is power and it uses USBC PD. I don't remember what this one is for. It might just be a USB 2.0 port or something like that. And then there's UART so that you can get serial data off of here if you need to. So, cool to have all those things inside. There's also M.2 connectors for more expansion. I think you can do Wi-Fi on here uh if you wanted to do that as well. Um, I haven't taken the board out because I in all my testing I wanted to make sure I didn't touch anything or mess anything up. Um, and at some point I will take this thing apart and and take a deeper look. And yeah, looking at the specs, that is a USB C 3.0 port right here. And then internally there's some more fan headers. It looks like there's even an edge connector here. I wonder if that's possibly for mini display port or something. I don't know. Uh, but yeah, cool box. And I will leave links to everything in the description. Um, happy 2026 and until next time, I'm Jeff Gerling.

HUGE thanks to Patrick from @ServeTheHomeVideo for helping with this test! Links to everything I mentioned: - Mono Gateway: https://mono.si - Tomaž Zaman on YouTube: @tomazzaman - ServeTheHome YouTube: @ServeTheHomeVideo - ServeTheHome Website: https://www.servethehome.com Tomaž sent me the dev kit Mono Gateway I am testing in this video, so I have marked the video as 'sponsored' (I also ordered one for myself, prior to Tomaž sending this test unit!). Mono did not offer any other form of payment, nor have any input into this video's contents. See my sponsorship policies here: https://github.com/geerlingguy/youtube?tab=readme-ov-file#sponsorships Support me on Patreon: https://www.patreon.com/geerlingguy Sponsor me on GitHub: https://github.com/sponsors/geerlingguy Merch: https://www.redshirtjeff.com 2nd Channel: https://www.youtube.com/@GeerlingEngineering 3rd Channel: https://www.youtube.com/@Level2Jeff Contents: 00:00 - What is this box? 00:58 - 51.2 Tbps in Phoenix 02:00 - The Million Dollar Network Testing Box 03:33 - From 1.6 Tbps to 10 Gbps 06:14 - SFP+ is always upside-down 06:58 - Saturating 20 Gbps with real traffic 10:13 - Real web traffic and latency results 11:55 - Firewall testing next year 13:39 - Mono Gateway works 14:53 - Specs and happy 2026!