 So we're here with EasyChip here at the Lunara Connect. So hello so who are you? I'm Barry Spinney. I work for EasyChip. EasyChip about nine months ago bought a company called Tylera that I work for. So a lot of the work I'm showing here is actually on the Tylera based platforms. Which Tylera are we looking at here? So what we're looking at here is a box called the Liberty Blue Tyler Stream. It's actually four completely independent boxes effectively. Each box has got its own disks and 40 410 gig ethernets in a Tile 36 chip. Tile 36. That means 36 CPUs of the Tyler architecture. What is a Tile 36? It's a single chip that has as I said has multiple memory controllers. It's a complete system on a chip with in addition to four 10100 gigs it has PCIe interfaces, multiple memory controllers and 36 completely parallel CPUs in the same chip. But it's not ARM. Our next generation chip called the Tile MX is very similar using a lot of similar technology but we're replacing our own CPU with an ARM chip. 64-bit? 64-bit ARM V8 with all the instructions all the SIMD instructions all of the crypto and neon extensions. How soon is that one coming out? It's gonna be the second half of next year. I don't know I don't have a date I'm not involved in that part of it. But this one is what kind of chip is it? It's not ARM? It's not XC6? The instruction set is based on something called the raw instruction set from RAW that was invented by MIT a bunch of years ago. So it is a proprietary instruction set that is used by Tileira only. It's got a lot of registers very heavy register centric. It's very nice architecture actually very very efficient. This sounds like a really powerful solution. It is. In fact I don't know when they at various times at various times we have had the highest CPU rating of any box any single chip in the world. In the world. In the world. We've actually had those records. I don't know today. When did they come out? The Tile 36, Tile GX 36, I don't know maybe two years ago? Two years ago. Something like that. You have been the fastest chip in the world. At the time we came out. And now you're gonna get over to ARM and 100 even faster. Well we don't know when we come out if what other people are doing but we are looking for 100. 100 cores. 100 ARM CPUs each running at you know a very high clock rate. So 100 cores. Yeah that's what you're saying. Yeah 100 cores and and in addition to the 100 CPUs it's all wired up using our mesh technology that we had in Tile GX but also it has all of our hardware assist from Tile GX and a whole bunch of new hardware assist. So in addition to the M-Pipe hardware hardware classifier, hardware scheduler, it has the crypto engines both public book and public key but also as hardware traffic manager, it has hardware counter assist, atomics, it has hardware TCAM, long as pre-fetched match hardware, hashing hardware. It's got just lots and lots of hardware assist for networking. That's that doesn't sound like an easy chip. It sounds like a really advanced one. Well actually there are actually we're now another easy chip we're actually kind of have two kinds of lines of chips and the traditional easy chip which has just been announced they just announced the NPS line which is a 400 megabit interfaces chip and it's actually even bigger than this chip's going to be. 400 megabit? 400 gigabit I'm sorry. 400 gigabit. 400 gigabit of capacity and that's you can buy you know we that's gone into a production or it's taped out so that's our highland line. What is that? Is that an arm or what is that? No that's actually using a CPU called an Argonaut Riscore which has been used by other people. It's a standard architecture but it's not an arm. So that one is really more targeted for very high-end routers and switches and very powerful things. It's actually previous generations of that chip actually were in a lot of the big routers that you currently can buy. You know it has massive traffic management capabilities, massive data movement capabilities. So the Telerra, the Tile chips GX and MX are the lower line, the lower line but they're more flexible because because we have an arm core which is more easier program but it doesn't go quite as fast. It's not a 400 gigabit system, it's a couple hundred gig system. A couple hundred gigabit system. Sounds crazy fast still. It does but you know people move fast so at the higher the NPS line they'll be looking at terabit systems very soon so we're going up to they're going up to that level and we're going up to the 200 gigabit level. Terabit is for very special projects right? Well again the backbone high-end routers and switches that we're coming up with in the next couple two three years. A lot of times these chips that we develop at EasyChip and previously Telerra they have to be designed in really early into these Cisco routers and Juniper and other devices. So what those companies are going to ship in three to four years they want us to have the chips develop now or the next year because they need another year or two to build software around it and make a product from our chips. So you know we have to be several years ahead of what these huge bandwidth are and that's what our customers are asking that they will need in the next couple years. And building chips is not something that's so quick to do to be done. No it takes a while. It takes several years. It can take anywhere from year and a half to three and a half years something like that could be anything that range. So the people in charge of Roadmap have a very nice vision. They know what's gonna happen in the future. They have to predict pretty well and to do that we are very we work very closely with our customers past customers and have them help us predict what they will need often for five years from now and then we work with their Roadmap to try to guess what our Roadmap should be. So you're going to make an arm process over the whole bunch of you said mesh you said some hardware. Mesh is just a wire connecting all the arms together into one unified whole. That's like the cache. It's like no it's cache coherency and also the interconnect that connects all the cores to common memories and common subsystems and accelerators so everything can be so it's common communication. That's what we call the mesh. And then you said some accelerators. You have a bunch of accelerators in there. Yes. And this is the special sauce to bring in to these. And what are you showing right here? So what I'm doing right here is I'm actually showing a traffic generator again running on one of these independent boxes. Two different traffic generators sending in traffic. Each one sends in 20 gig over two 10 gig ports into a another box running my latest ODP implementation. So I just recently wrote a Tile GX ODP native implementation. By native I mean that it's from not a port of something else. It's built from the ground up to work with Tile GX hardware as efficiently as possible. It's not just reporting slow speed software running on of generic Linux. So what I'm showing here is that with my implementation running a standard application ODP application like either L2 forward application or OBS in L2 mode I'm showing I'm trying to show that I can get 40 gig in and 40 gig out through my ODP real ODP application running on top of my ODP code. So the ODP application and the ODP in general is kind of like this new movement, right? Yeah. And this is a big deal. It's changing up and it's it's working. It's useful. That's what I'm well I don't know if it's it's not production yet. I mean I'm just showing a demo of it because it's not fully implemented but I'm showing that the key point is A we actually now have a real ODP implementation doing real ODP running real applications and B that runs very high speed and C that it's not just a port from some a quick port from somebody else it's built from the ground up for our hardware optimized to optimize to our hardware and built around our hardware so it's a real it has the potential to become our production quality real shipping ODP implementation for Tudjx. How hard is it to make software solutions that are crazy optimized for hardware? It's hard but it's basically what I love to do that's my specialty that's my secret sauce or that's what I try try to focus my career on. I always like to say you know if something needs to go really fast on any hardware whether it be specialized or not that's a project that I personally love to work on. I don't want to be porting you know lower speed applications on generic Intel box or anything that's just not my interest. So you don't like you don't like bloatware you don't like uh no I want things go fast I want super optimized yes I love doing fast things. Do you like arm? Oh yeah yeah it's nice so so this implementation will be moved over to our arm chip as time progresses once we you know so it's it's not just going to stay on GX but we have to change you know some numbers of the things of it to to make it work on the max. And some things will run on the arm and some other things on the hardware accelerators and uh yep and that's my job to figure out that split and you know we always tell our customers and how the world how we split it but obviously we want to make sure our hardware is getting its maximum visibility and benefit for the customer. So what have you been doing in the last a new career what what what do you do? A lot of things um going way way way back I was involved at deck at the um something that probably nobody remembers something called the giga switch which at the time was I was one of the principal architects I was just 36 100 megabit per second fddi connections into a single box and that box was used in fact at some point of like I think half of the traffic in the eastern or western us went through that box really and one box one through one box when was that oh around I know I gotta remember my time scales here give me one moment it was I guess 94 96 something in that range beginning of the web right oh yeah it was early and we were we were the only game in town though our competitors only had you know either you know two or three ports and we had 36 the 30 40 36 we had one box running half of the U.S. internet through the back through the uh the core of the network yeah that's what I was told I never actually visited it but this box was very very high what was the company you were that was digital corporation deck big big company obviously deck so I worked in the deck advanced development group for many years and doing that project and it was called a giga switch and it was revolution it's time and then later I went and worked at some stuff for ATM for four systems and then I spent about 15 years doing my own startup which was doing security denial service all sorts of security things denial service at the end IPS IDS eventually it got bought by some European company they took it a different way and what's the name of your startup it was called top layer networks all right so uh and here you have you been to the narrow connect before last year I was here so what do you think about these narrow people nice they're very bright bunch and uh there's some discussions going on the things are moving fast or yeah they're moving fast I could always you could always like the things move anytime you have a consensual process and an open process like this is the community process it's going to move slower than what I'm used to where I come from a world where it's not as consensual it's not open it's not as um you know community-based so you know you basically decide everything well sometimes I've been on bigger projects where we do have to argue about it but it's all within the same group so you can move faster that way so so it's a little slower than I'm used to but on the other hand it's moving I think as fast as hopefully it will keep how can it move faster I don't know that's what I'm saying when you have open source consensual community-based projects they're just going to they just have to move slower than a proprietary internal company only thing to do can we see some stuff on the screen yeah let's see what I have here I'm not sure what I have because I was running it before and what I'm showing here is it's not that interesting so basically I kick off the traffic generator the two traffic generators and I kick off the odp code I'm not sure what the odp code looks like so when I run the only um I'm running yes I'm running something else so here's a typical run so this is running um running this is how when I run the all-2-4 application it's what it puts out every line puts out how many packs per second max number of packets drop or it puts out every second and that number it's just a packed max packet count so what I do is I correlate it by looking at the my stats so this is an example stats um so what this program is this program is something called m-pipe stats and what it does it shows for each of these ports x to be one two three four which are the four ports here it shows the total amount of packets being um transmitted the number of megabits transmitted and the amount of packets received and gigabits received so you can see it starting up here you can both traffic generators and then it hits its peak performance here if you look at this range they're all the same numbers it's all basically 9.9 gig 9.9 gig received transmitted and about 800 thousand that's per interface so that proving that we're really getting this performance and I'm not losing anything and I'm really keeping up with this traffic level so um in previous shows we've actually had uh a GUI back in that actually had meters and showed me things it was more graphically rich for this show we I just didn't have time I was doing it all myself and I just didn't have time to do anything graphically richer than this all right so there it is uh faster internet it's really important because people want more bandwidth and uh ODP and uh Leonardo LNG and your new crazy pass chips are going to make that possible otherwise yes otherwise uh there might be a bottleneck and just nobody's going to be able to stream 4k videos exactly it's going to be an issue but hopefully not all right hopefully we'll be there we'll be ready for when the industry needs us