I'm blown away by the work done for Meta's Evenstar team. I loved the talk from XRComm at a OpenCompute Global Summit, in support of adding Evenstar to OCP. (The slides in particular show lots of figures-of-merit that evoked a strong response for me.) https://youtu.be/SXNH3ddpv1k
The hardware looks absurdly economic & performant, crazy modular, and there's amazing changes happening on the software side too with much more being done closer to the edge in converged hardware.
It's probably the highest tech (and extremely cost optimized) open source hardware design ever. It looks staggeringly far ahead of commercial offerings. If this doesn't jumpstart massive change, shame on the world. Modular designs for: power amplifier, power supply, digital control board, resonant cavity duplexer filter, signal processing firmwares, heatsink/case.
Right now we are deep into the days of cloud. It's hard to imagine people buying servers! So it's definitely hard to see what role OCP might fit into, with that perspective.
It also feels somewhat like you have your ask inverted:
> I'm interested in how you could build a business around it.
Around or on top of? OCP is kind of there to make suppliers job much harder; OCP is here to twist arms of people making hardware. OCP exists to remove special sauce, to stop vendors from trying to get competitive advantage. Anyone making hardware generally is under threat from OCP, as a kind of ruthless buyers cartel who know exactly what they want, and who is happy to buy the specified product from anyone.
If you are trying to be further up the stack, OCP is amazing. There's a clear ask, that you and many other buyers are hopefully all asking for. There's systematic design efficiencies throughout; instead of only high end special gear having rack-based power (or mini-me edition: large blade chassis, those chassis requiring tons of extra/expensive metal/backplane), it's just standard, normative design. Higher efficiency and cost-optimizing.
Aside from big obvious racks... there's standardized bios/management interfaces that have become vastly more prevalent. OCP nics show up on a very sizable amount of server gear. The buyers have had no real concerted effort to say what it is they want until now, and OCP is a communication/coordination channel for buyers to figure out what they want. It makes the industry better, and the evidence so far has been great, if slow.
With any hope, we are shifting back to an age where more people & more companies actually run their own hardware. I personally see really amazing progress from places like Kubernetes as a key enabler, a tenable start towards a data-center OS, where open source services like cloudnativepg can begin to emerge & compete with the expensive well-served SaaS world we've become addicted to (beat back some Service as a Software Substitute (SaaSS)).
> Facebook's Tioga Pass the innovations in that system are simply not available for any price to the enterprise buyer
Seems like OCP may not be as open as appears, when some 'secrete sauce' is behind FB wall. Seems like better to go with Oxide instead of this OCP approach. Disappointing, but not surprising coming from FB or any of FANNG group.
It's not so much an "FB Wall" as much as OCP vendors not willing to engage with small clients. Big clients can leverage their size to acquire supply contracts from ODM's like Quanta that quite often would just ignore smaller companies.
Arguably the big innovation of Oxide is providing a turnkey system in the same range that is comparatively easier to buy.
OCP and Oxide two different but not exclusive approaches.
There are vendors providing different levels of integration, including groups that will ship you turneky systems akin to Oxide (or AWS Outpost). Those might depend on OCP Standards to provide higher level of integration than integrating normal HW solutions. Starts at integrating somewhat individual components, going through fill racks, to entire data center standards or designs for Modular Data Centers which you can load onto a truck, deliver to location, hook up and have them working (some might remember Sun containerised datacenter thing, which arguably reminds me of bigger Oxide rack in some ways)
Similarly some bits designed for OCP (or hosted by OCP) contribute to wider environment, like OpenBMC, OCP Mezzanine cards
That's exactly right. OCP is an important effort because it encourages all of these things. (And that it has been more encouraging of software in the last few years is a step in the right direction -- but it is still broadly disappointing on the open firmware front.) Another concrete example to add to the ones you listed: we are using the presence of the OCP DC-SCM connector in next-gen AMD dev boards to be able to have our service processor be able to control many aspects of the SoC before we have our own boards taped out.
All of that said: it remains true that going ODM-direct really isn't possible for all but the largest buyers -- and even then, it is a big upfront investment in terms of both capital and engineering time. (And one remains entirely on one's own for software!) So while we are appreciative and supportive of OCP, we also very much believe that there's room for the kind of turnkey solution that Oxide offers.
OpenRAN, or concept of RAN as Software isn't exactly new. If anything most of the new 3G / 4G / 5G proposal are first tested in software simulated world before getting the efficiency from specialised silicon solution from the likes of Ericsson and Nokia. ( or Huawei ). What is new with 5G is that we could move most of the processing originally happening on site, now to a centralised backend sever. ( Something I have been saying on HN for 5+ years about 5G being cost reduction for MNOs ) You basically have Antenna and Fabric connection and the workload being done in some Datacenter fairly close to the site. A site within 100Km radius would add less than 1ms in network latency. This simplifies a lot of the on Site installation with Power Requirement / Limitations. Once you add Massive MIMO on site you quickly ran into power budget issues.
This centralisation doesn't always means open, or at least software defined. Because the efficiency is easily 100x plus comparing general purpose CPU to purposed build solutions. But interestingly enough we are now at a stage where we have widely available GPGPU and FPGA solutions which changes the equation. Consider  GPU-Accelerated 5G Open RAN from Nvidia. Since all the purposed build solutions from the likes of Nokia and Ericsson are never Fabbed on the latest Node, GPGPU while not being as fast but are fabbed with Node 2 generation ahead and can be replaced every now and then for upgrade and repurpose older generation for something else. And considering you can now rent these from the likes of AWS and Azure. It dynamically changes the OPEX cost calculation.
I've always had the following idea about cell networks and I am curious what people here think of it:
We could mandate that providers must accept cell phones which can dynamically switch between networks at will. The wireless protocol can communicate the cost of access at any given moment and the device can decide which network to use.
Under such a system, lock-in is eliminated and providers finally see the full power of competition. Right now with high entrance and exit fees, competition is reduced significantly. This would aim to fix that.
I think Verizon was required by law to allow 3rd party devices to use its acquired spectrum (I think it was the ~900Mhz?), and they just dragged their feet indefinitely under the guise of safety and security.
Unless the consequences of violating the rules is sufficiently expensive to harm earnings and growth, any rule you set forth is just cost of business.
i.e Local Roaming. The problem is there is no way to ensure QoS over the network. You will need additional functions to ensure the network is best every time you switch.
>What do people here think of that? Would it help?
I am assuming you are talking about US MNOs or Carriers. And I am often lost and dont understand how anything in US works. It seems there are decent options, like Mint  which offer very decent 5G price plan, especially considering the wages across US. $30 for Unlimited or $25 for 20GB is very respectable.
So I am not exactly what the proposal would help, or trying to fix or improve.
Edit: I will add more context. If it is about better coverage and more site being built. Then this is actually a property ( surprise surprise ), regulation and market issues. No Technology is going to save you if the market, regulation or the people / culture does not allow you to install a 5G radio because they are scared.
You will end up with something like energy demand response market. Where providers aren’t incentivized to build out the network for redundancy or QoS. Then you would need the Govt to step in to provide further incentives to build in rural areas where it’s uneconomical to build network. Naturally over time you will still have monopolies in certain areas because building the network is still costly, in the end winner takes it all. But those monopolies will be regional.
There would be no more incentive to build the widest coverage. In cities the fastest or cheapest network would win the spot market. Without the ability to reserve a wider network to their own customers whom they can charge a higher monthly bill, the additional CAPEX for rural towers doesn’t pencil out. To make rare expensive services like rural towers and generator backup work you’d have to charge a usurious amount per MB or session, or enforce a minimum purchase length to avoid adverse selection. And then you’re back to contracts.
I'm very confused - isn't this how phones work right now? Sans the dynamic cost marketplace of course - usually the phone sticks to the network of the sim card you've currently got slotted in, except when Roam-Like-At-Home kicks in. But you can manually select whichever network you want in theory.
European here and not an expert on this, maybe it's different in the US or I'm gravely misunderstanding something.
You can't select any network, your home carrier has to have an agreement with the network you select.
They're talking about switching operators, not roaming - roaming means your traffic is still routed to your own provider. The concept, if I understand it correctly, is that you'd pay the best offered rate per byte around you at the moment, and your phone would be dynamically switching networks based on the offers.
The problem with this is that unlimited plans would go out of the Windows and we'd be back to paying per byte. Not great imho.
Entry and exit fees?? Is that what it's like in the US? In Australia you only get locked in by bundling the hardware on a contract, but if you own the phone, you can switch between providers as often as you like.
Ostensibly there are not fees for switching carriers, but there is always some random “setup fee” type thing that is unavoidable.
I think the parent here is suggesting though that you should be able to dynamically choose which tower/provider your phone is using, and would just be charged for the time on the system that you use. It’s a bold idea, and would destroy the existing carriers business model (which is largely dependent on contracts)
Isn't it how MVNOs worked, even in 3G? The phone tries PLMNs listed in SIM, the Node B/RNC relays auth requests to HLR that corresponds to that IMSI or number, and that HLR could be anywhere so long the RNC knows the ... address?
The SIM card contains information that allows it to negotiate with available cells regarding which cells and with what parameters it can connect.
The cell relay network (NodeB/RNC/etc) the connection to their owner service which sets up authentication with the SIM card - either by consulting service's "phone number database" based on the IMSI number of the SIM card, or by forwarding it to roaming peer's database (HLR). Then it keeps tunnels between phone and appropriate gateway.
SIM provides the identity of the "account/connection".
IMEI provides the identity of the actual terminal (arguably a bit like MAC address) - called "User Equipment" in UMTS/LTE/5G
BTS/NodeB/RNC/etc. - the radio segment - BTS/NodeB/ENodeB being the "base stations" and RNC is the radio controller (also called Mobile Management Entity in newer standards) which handles setting up connectivity from UE over NodeB to appropriate gateways (connecting to IP network or circuit switched phone networks)
HLR/HSS - Home Location Register / Home Subscriber Server - the part of the system that authenticates the subscriber identified by SIM. It's what maps the connection to right entity in network (to route phone numbers, match billing, etc.) - when you use roaming, the RNC/MME is set up in a way where it can find your original HLR/HSS (tackling extra fees as necessary)
gateways - different depending on version of network as well as whether you use IMS-only network (aka VoLTE/VoNR, in actuality part of 3G known better to people here as VoIP/SIP), or whether circuit switched fallback is used (operates purely on 2G and 3G, with latter being added in later revision).
The issue is with billing and which mobile provider is responsible for the phone number. If your phone is dynamically switching between 5 carriers, does that mean you now need to have 5 billing account as well?
Without changing existing protocols, it would work like MVNOs (mobile networks without own networks) operate, where they provide or lease identity management services and gateways to rest of the phone network, with roaming links set with owners of actual physical networks
Ah, OpenRAN. Germany's up-and-coming new fourth provider 1&1 is exclusively aiming for OpenRAN , with the antennas (at the moment, supplied by Japanese Rakuten) being just antennas (i.e. modems + amplifiers) and everything else be done in larger area controller data centers linked via fibre and everything being done in software .
It will certainly be worth the popcorn to watch that one from the side, personally I hope for a success of OpenRAN simply due to my dislike for vendor lock-in.
* Huawei sells better and cheaper network hardware than anyone else.
* The US has previously kept Huawei's market share at bay using every political trick in the book, but the US's reach only goes so far, and other nations are buying large amounts of Huawei hardware, making it even better and cheaper.
* Modern mobile networks work far better if you buy all the gear from one company. That makes it even harder to persuade companies not to use huawei gear. It also makes it even harder for competitors, who now have dwindling sales, to keep up.
As someone who’s seen the code for Qualcomm’s cellular modem (granted in 2009) and worked with their engineers, it would be impressive if it was tangible worse.
I think we should be clear about the strategic threat of having a foreign company own telecommunication infrastructure, but that’s technically true for US allies as well. The main distinction is that China is not an ally.
Canada actually has rules about telecom companies needing to be owned by Canadians but I don’t think that’s served them all that well (& doesn’t touch the question of who owns the infrastructure).
1: Simply bringing it down. If tensions escalate and someone has the ability to cripple their adversary's communications, obviously it's a string they can only pull once, but knowing they can pull it when the time is right, is a big deal.
2: E2EE doesn't insulate you from baseband vulnerabilities. You can think of a modern phone as having two processors, though in practice it's many more, think of the Application Processor (AP) and the Baseband (BB). The AP is basically the CPU that user code runs on. The baseband (so called because it deals with the radio signals once they've been downconverted from their broadband modulations) is sort of like a NIC, it provides connectivity to the main CPU.
Now, if you're familiar with server NICs, imagine a server BMC that has god-level access to the hardware on which the application runs. That's roughly how the baseband is situated in the phone.
Obviously you'd only connect such a BMC to an internal management network that's only accessible to trusted parties, right? Not to one that your adversary controls?
YEahhhhh. If the adversary controls the cellular network and talks directly to the BB/BMC, it doesn't matter how secure the AP/CPU is, it's game over.
>2: E2EE doesn't insulate you from baseband vulnerabilities. You can think of a modern phone as having two processors, though in practice it's many more, think of the Application Processor (AP) and the Baseband (BB). The AP is basically the CPU that user code runs on. The baseband (so called because it deals with the radio signals once they've been downconverted from their broadband modulations) is sort of like a NIC, it provides connectivity to the main CPU.
The situation is somewhat better now. iPhone's modems are separated from the main CPU and communicates with it using some sort of serial link, and AFAIK tensor based pixels have modems isolated using IOMMU.
To be fair, the discussion here is not about baseband processors afaik but the infra that the baseband talks to. Of course you could use that as a hop point to infect all attached basebands because they suffer from a software monoculture and you could get a lot of them. Other less obvious attacks would be selectively downgrading the protocol phones can see the tower using to remove the encryption capabilities. This is similar to what Stingray base stations do.
Exactly. Control of the management/cellular network is not _precisely equivalent to_ control of every machine/phone attached to it, you still have to attack the BMCs/BBs.
But I think it's safe to assume there's always a BMC/BB vuln, especially if the firmware development for it takes place in that same adversary's country. Not having the adversary on your management network in the first place is probably sane.
Firstly E2E doesn’t protect the metadata. So a compromised backbone would still give you insight into who’s talking to whom which is what the NSA doubled down instead of trying to sift through raw data (they still do but social network monitoring is easier and can give more insights).
Secondly, if you made the infrastructure, you’re more likely to have attack capabilities that can take it down more easily. Probably helped US and Israel that Iran was using tech from Siemens which was easy for them to acquire and work on since it’s from an ally country. In the fog of war, being able to sow chaos into the home front is useful for disrupting the war operation. The actual message content is probably less valuable.
But the truth is we don’t actually know. There hasn’t been a war between major powers in the telecom age. From that perspective, it’s wiser to keep infrastructure decoupled to avoid surprises and mitigate risk because it looks like conflict may come sooner rather than later given how both sides have been building up and conflicts have been escalating globally.
Sure - but security isn't the major focus for many buyers. As long as it routes data from A to B and can keep the population connected, many governments and companies are happy to take the increased spy-risk in return for lower costs.
Lets be honest, the future is end-2-end encryption and a dumb network, and then the security of the network becomes far less important.
There's not a network out there that won't work smoothly without support. Hardware breaks, software breaks, vulnerabilities are found. When sanctions ratchet up or bombs start falling, which support contract is still honored?
> When sanctions ratchet up or bombs start falling, which support contract is still honored?
The contract with the company owned by the right honourable member of the house of lords of course. Not only will they fulfill the contract at the highest possible price, but they will fulfill it with the cheapest, least compliant product possible and worry about being sued later ...
It's akin to everyone in a company using outlook and an exchange server, vs everyone using a variety of mail clients and a mixture of server software.
A bunch of features only work well if everyone uses outlook (integration with address books, calendars, free/busy info, auto video call invites, recalling messages). And you're gonna have far more headaches if you try to run a mis-mash of different servers and clients - although it is technically possible to get it all working.
But that’s the point of standardization. You are trying to ensure that the core functionality works for everyone regardless of who the vendor is. A mobile network much more akin to SMTP/IMAP than it is to intra office calendaring (but even then, CalDAV might have an opinion to offer).
But this interoperability requires standards and vendors to adhere to those standards. But the only way you can get true interoperability is to test it, which is what this funding is about (IIUC).
Absolutely. Part of the challenge is that Open RAN opens up more (and new) interfaces that have to "play nice" and interoperate.
Another part of the challenge is the commercial business models part - monolithic RAN vendors can integrate vertically and horizontally to offer a fully tested suite that goes beyond just what's in the standards. To get like-for-like performance in things that matter, you often need more than the base standards. For example there's all manner of proprietary data exchanged between radios to coordinate performance and interference, which is vendor proprietary. That's needed to win an edge over others and win contracts, in the eyes of big vendors.
The issue right now is that even just getting 2 standardised boxes to work together needs exactly that testing you refer to - different versions of ASN.1 schemas for example, can break interop. There's often vendor tweaks to edge cases to fix interop.
In the old world of a single vendor networks there was some "goods out" testing carried out on equipment to ensure it worked together as a solution. When you buy components and put them together, you need to do that testing (hence labs like this).
One final interesting challenge will be around funding future R&D. Big telecoms vendors need to invest many billions per "G" to keep up and get patents into standards. Open RAN is likely to drive disaggregation and cost savings for operators, which will reduce margin and "fat" in contracts. That might make it harder for innovation in the "next big thing", if they can't make a profit from charging a margin on some COTS servers or software platform. Which isn't to say that more competition is a bad thing, but just that disaggregation might cause a slow-down in future innovation if reduced margins can't sustain it. Operators want Open RAN for cost savings, and that means someone's margin needs squeezed.
This is for the carrier side of the equation. Even if a carrier uses only one vendor for all their cell tower RAN equipment, they would need to support whatever phones end-users use e.g. Samsung/Apple/etc. It's more like Netflix deciding they'll use Apache on x86 for all their web servers but still accepting standard HTTP connections from Firefox/Safari/Edge/Chrome/etc on x86/ARM/etc as user agents. This is more sensible than you portray and is common in other industries so not sure why OP would think mobile networks are an exception to this rule.
I don't understand the /s tag here. Both statements are categorically true; These facts just have consequences we don't like, and somehow must go out of our way to counteract if we care about our ideals.
My statement, at least, is categorically untrue. I had the displeasure of building for the web in the time before Firefox became mainstream. IE6 was essentially the only option. And it sucked! No competition meant bugs didn't get fixed. The web was IE and Microsoft couldn't give a flying fuck whether it followed the standard let alone behaved consistently even between page loads of the same site.
But no matter how good or bad a particular browser is, you still have less work to do if you only care about users on a single browser (compared to that browser _and_ other browsers). Even if the other N-1 browsers have no quirks relative to the first one, that's still more testing to do and support to give. It's objectively easier to target fewer platforms, which is why you see the laziest web developers put up warnings (or even block requests) for clients with the "wrong" User-Agent strings.
>* Huawei sells better and cheaper network hardware than anyone else.
If you spend a decade spending 10s of millions of dollars (or more) on R&D into cellular networks, and then I just take all of your findings and build a competing product, do you think I can sell my product for less money?
If we assume all network gear is backdoored (it isn't but let's just assume it is because I know that will be the retort) would you rather the backdoor be controlled by a friendly, democratic nation, or an openly hostile, expansionist, quasi-communist/quasi-dicatorship?
If you answer both of those questions truthfully, on what planet and by what logic would you suggest the world standardize on Huawei networking gear?
> If you spend a decade spending 10s of millions of dollars (or more) on R&D into cellular networks, and then I just take all of your findings and build a competing product, do you think I can sell my product for less money?
This argument would be more convincing if the others had a competing product first, but it seems like they don't.
>This argument would be more convincing if the others had a competing product first, but it seems like they don't.
They literally stole all of Nortel's IP, and then hired away their engineers once they had employee data and knew who was working on 5G. But other than that, I guess nobody "had a competing product first".
>Its how the USA was built, seems a bit unfair to hold China to a different standard.
You mean it seems unfair to hold China to the same standard in 2023 that every other developed nation is held to? Instead, they should get to rewind to the 1800s? Should we turn a blind eye to what's happening to the uighurs too, because Native Americans?
All of that is ignoring the IP theft that occurred early on in US history isn't anything even remotely resembling what China has done. You're comparing someone having a bonfire to the Canadian wildfires and acting as though they're the same magnitude.