 Serum menu, ampak. Ponočenek, ponemečenek Serum menu, zakres. Je tertekčne delniko, kako se možete odvniti prav m настолько in prišlo lim barkin. Wi so ma preke slasvene dolge izbo tardim v malo začna prince kr vaccine za generations taz zač Interesting mandor. Daj se počečenje očečnja začno tega, tko tudi je tudi pripravil, ali pripravil je pripravil, pripravil je, da so pripravil všeč spolaj, ali neč našel, ali pripravil je našel, da bomo počečenje. Zato pripravil je. A zato, da je bilo pripravil, je to pripravil vseh komplek, kaj je, da je pripravil vseh komplek, na nekaj leblj, doči je to skupnost. Poči je to mene? So svoje gdje je vsožite, da je dobrodo power, pa odočim... Znamenj, daj telo je občin, skupne zvuk, so mene šta, že se spročila, da je tukaj je zvumil vzad, tega brzina, kako je se vzpušila. E.g. rehter, sa jistim, zato je pravda, če nekaj poživuje in občajne je? Ne zelo. Tako če nekaj, da naredem vseznačnji projekto, poradno, da bi smo počajna vseznačnji vseznečnji subplikacija, zato da bi se odčeli kod na hodnih hodnih, pa nekaj, da bi smo počajni vseznačnji vseznačnji vseznačni. The same with configuration tool. So what will be then, first of all, we need to have CxL switch as the real stuff that we can test. But we need to have also some infrastructure to manage to use, and so on. So this is something like really, really first step. It would probably be rolled into a project like OpenBMC. Pače nrga, je to, da je vstavljen, v ter velikom tom enovavu se odらčilo, da je razselje na ampak razvalovješčenih zap hurdleja, tko je je zmizirana, kako ta ne gav tudi je tukaj, da je to tukaj eliminates in nekaj rejner, kaj da je od vrstavna zelo, na kakva razvala, But in a data center type environment, it's probably going to be running on a BMC in a memory appliance, rack mount of memory appliance. And the thought was, yes, it would expose a red fish interface to the outside world and there would be some open source project that translates the red fish into the CCI level commands. I wonder giving the state of where the rest of the CXL management conversations are going, if that winds up being just an RDE pass through basically. So the FM isn't actually doing any translation at all because that brings with it the maintenance burden of keeping on top of the red fish mapping and the CXL spec. There's like an RDE requirement pushed down to the switch or the multi-headed device. So that, I mean, at that point, I guess you could say that the embedded logic translating that RDE into actual device management operations within the switch or the multi-headed device, that's the fabric manager. But I don't know, but the idea is that the real intelligence of the system is in the orchestrator, it's not in the fabric manager, it's really just a translation layer. And, yeah, it will need to be a distributed approach, I think, definitely if you're talking about racks of hardware. But to the point of getting an open source project off the ground and getting a center of gravity going, like focusing on QAMU, focusing on the aspects of it that we can do in a direct attached model first, because I think DCD is going to be direct attached first. And switches, I don't know when this is going to happen, but at least, me as a developer wanting to test kernel drivers and Linux interactions, I'd want to use an open source tool, and then maybe that gets you your center of gravity. If people are using this for development and somebody wants to do proprietary extension, maybe they'll use this as their baseline, or they'll say, wait, we don't even need to be proprietary here, just put it into the common FM, because, yeah. I would hope that there's nothing super value out about an FM, it's just moving the memory around the architecture, so. Excuse me, so basically we are working on a similar thing right now, the differences we are writing in Go language. The question is, for the Refish implementation part, there are specifications from GMTF, but it does not clearly, like, I cannot understand if it's like, when we want to use the orchestrator to talk to the Refish, do we expect to send a pass to compose a system using resource blocks? Or do we directly find those mailbox commands, map the resource? So I don't understand Redfish either, so completely agree. Yeah, I mean, so GMTF has published a mapping table, basically translating the CXL management commands and properties into Redfish elements and attributes, but that's more from the context of individual endpoint discovery and management, which is, first of all, I think that's the biggest first step, right? Because even without dynamic opposability, even just from an inventory management through identification perspective, you're going to want to be able to pull the serial numbers models, all of that, get a picture of what's running and health from that before you start dynamically moving things around. I will say, on the question of the switch, I don't think anyone's going to get to play around with the CXL switch till 2025, 2026, but there are already demos on YouTube of multi-headed devices, and Tans and I, they're part of Marvell now, but they actually had a YouTube up of dynamic capacity device on a multi-headed device for a while now. An FPGA type system doing pooling on a multi-headed device, I think that'll be the first place to look at this. I would also say that the OCP Composable Memory Systems Group, they're working on exactly what you're talking about, developing a POC system, a reference architecture and an open source project to release to kickstart the industry on that. So, I can share with you some resources there. I'd encourage you to participate there. Can I add a comment? Yeah, go ahead. Okay, John, go to the mic right now. Just FYI, the Open Fabrics Alliance is also doing some fabric manager work. It's pretty primitive, but it's open, and it involves a redfish discovery and representations and things like that, and it's basically warming over Gen Z stuff, Gen Z fabric management. But anybody who wants to know what all is going on in this area, I don't know about that, too. Thanks, John. So, now, in this one session we've erupted three independent implementations. I don't know if there's any forum for these groups to work together, or are we just going to injury in a world where everybody's going to have their own FM? I would think that OCP would probably be the best place for an open fabric manager implementation to come out. I would expect from OFA something more orchestrator-based. Yeah, that's what I'm thinking. The CXL consortium in DMTF are just going to push out specs and implementation notes and white papers and stuff like that. The actual work, I think, would be done by those groups. Okay, thank you. I think it will be great if I receive all this information. Okay, so thank you for the discussion. That's everything that I have.