ARM Joins Linux Foundation's 'Open Programmable Infrastructure' Project (linuxfoundation.org) 18
ARM has joined the Linux Foundation's Open Programmable Infrastructure project, "a community-driven initiative focused on creating a standards-based open ecosystem for next-generation architectures and frameworks" based on programmable processor technologies like DPUs (Data Processing Units) and IPUs (Infrastructure Processing Units).
From the Linux Foundation's announcement: Launched in June 2021 under the Linux Foundation, the project is focused on utilizing open software and standards, as well as frameworks and toolkits, to enable the rapid adoption of DPUs. Arm joins other premier members including Dell Technologies, F5, Intel, Keysight Technologies, Marvell, Nvidia, Red Hat, Tencent, and ZTE. These member companies work together to create an ecosystem of blueprints and standards to ensure that compliant DPUs work with any server.
DPUs are used today to accelerate networking, security, and storage tasks. In addition to performance benefits, DPUs help improve data center security by providing physical isolation for running infrastructure tasks. DPUs also help to reduce latency and improve performance for applications that require real-time data processing. As DPUs create a logical split between infrastructure compute and client applications, the manageability of workloads within different development and management teams is streamlined.
"Arm has been contributing to the OPI Project for a while now," said Kris Murphy, Chair of the OPI Project Governing Board and Senior Principal Software Engineer at Red Hat. "Now, as a premier member, we are excited that they're bringing their leadership to the Governing Board and expertise to the technical steering committee and working groups. Their participation will help to ensure that the DPU components are optimized for programmable infrastructure solutions."
"Across network, storage, and security applications, DPUs are already proving the power efficiency and capex benefits of specialized processing technology," said Marc Meunier, director of ecosystem development, Infrastructure Line of Business, Arm and member of OPI Governing Board. "As a premier member of the OPI project, we look forward to contributing our expertise in heterogeneous computing and working with other leaders in the industry to create solution blueprints and standards that pave the way for successful deployments."
"The DPU market offers an opportunity for us to change how infrastructure services can be deployed and managed," Arpit Joshipura, General Manager, Networking, Edge, and IoT, the Linux Foundation. "With collaboration across software and hardware vendors representing silicon devices and the entire DPU software stack, the OPI Project is creating an open ecosystem for next generation data centers, private clouds, and edge deployments."
From the Linux Foundation's announcement: Launched in June 2021 under the Linux Foundation, the project is focused on utilizing open software and standards, as well as frameworks and toolkits, to enable the rapid adoption of DPUs. Arm joins other premier members including Dell Technologies, F5, Intel, Keysight Technologies, Marvell, Nvidia, Red Hat, Tencent, and ZTE. These member companies work together to create an ecosystem of blueprints and standards to ensure that compliant DPUs work with any server.
DPUs are used today to accelerate networking, security, and storage tasks. In addition to performance benefits, DPUs help improve data center security by providing physical isolation for running infrastructure tasks. DPUs also help to reduce latency and improve performance for applications that require real-time data processing. As DPUs create a logical split between infrastructure compute and client applications, the manageability of workloads within different development and management teams is streamlined.
"Arm has been contributing to the OPI Project for a while now," said Kris Murphy, Chair of the OPI Project Governing Board and Senior Principal Software Engineer at Red Hat. "Now, as a premier member, we are excited that they're bringing their leadership to the Governing Board and expertise to the technical steering committee and working groups. Their participation will help to ensure that the DPU components are optimized for programmable infrastructure solutions."
"Across network, storage, and security applications, DPUs are already proving the power efficiency and capex benefits of specialized processing technology," said Marc Meunier, director of ecosystem development, Infrastructure Line of Business, Arm and member of OPI Governing Board. "As a premier member of the OPI project, we look forward to contributing our expertise in heterogeneous computing and working with other leaders in the industry to create solution blueprints and standards that pave the way for successful deployments."
"The DPU market offers an opportunity for us to change how infrastructure services can be deployed and managed," Arpit Joshipura, General Manager, Networking, Edge, and IoT, the Linux Foundation. "With collaboration across software and hardware vendors representing silicon devices and the entire DPU software stack, the OPI Project is creating an open ecosystem for next generation data centers, private clouds, and edge deployments."
This wont matter as long as Android lives. (Score:3)
Re: (Score:3)
Yeah, aside from buying purpose built hardware such as the PinePhone or the forthcoming Risc-V handset from Sipeed, there aren't many options.
postmarketOS has gotten several repurposed handsets to the usable as a daily driver phase but good luck finding one of those in a working state on the second hand market.
Doesn't Google have a 'free' day once a fortnight where they can work on their own skunkworks project? Not that any Google employees in the phone division read this website but consider contributing p
Linux phones (Score:5, Interesting)
Yeah, aside from buying purpose built hardware such as the PinePhone or the forthcoming Risc-V handset from Sipeed, there aren't many options
Another option (which is the route that the people at Jolla use for Sailfish OS) is company who make Android phone, but have a community support program that makes unlockable boot loaders, provides kernel source + driver blobs, and assist 3rd parties into leveraging those for whatever they need.
Sony's "Open devices" is such a prominent one (but Jolla partners with a few other smaller players).
But instead of just making Android remixes like everyone else, Jolla use their libhybris abstraction and runs a full blown GNU/Linux distro on top of the drivers normally designed for Android.
Ubuntu Touch used to do a similar approach (using the same tech by Jolla).
Re: (Score:2)
I always though "Jolla" was a terrible name, as it means "Begging" or "Choking" in Korean.
Languages (Score:2)
I always though "Jolla" was a terrible name, as it means "Begging" or "Choking" in Korean.
And it's a type of boat in Finnish (and several other European languages), used as pun answering an internal memo from back when Nokia got Stephan Eloped by Microsoft.
Whatever word you pick for a name, if short enough to be usable as a name, has a certain chance to collide with something inapproriate in a completely unrelated language at the other side of the planet. So good luck naming your project.
(Unless you're a large corporation and can spend quite some budget just making sure that your chosen name isn
Re: (Score:1)
but good luck finding one of those in a working state on the second hand market.
I regularly use https://swappa.com/ [swappa.com] to buy Google Pixel 4+ unlocked handsets (don't buy a carrier version, they lock the bootloader) which are carrier and bootloader unlocked. They work with https://lineageos.org/ [lineageos.org] and https://calyxos.org/ [calyxos.org] and my phone works with Verizon and Visible in the US. So far, I've had one Android app not work on it, so that's 99% success using it as a smartphone for everything over the past year.
Learn how to back up your phone; the only problem with second hand devices is not knowin
Re: (Score:3)
You could in theory replicate what happened with the IBM PC compatibles in the phone market and kill the closed hardware ecossystem
Re: (Score:1)
Re: (Score:3)
Re: (Score:1, Flamebait)
We need a mobile OS that can install on phones like Windows or Linux does on a desktop. Google does not seem at all interested in making that happen. Hardware without software to match is worthless.
You don't seem that the entire industry is killing you owning your hardware, aka they are killing plaintext microcontrollers.
Nope, they are locking down the PC, this has been a 20 year project to kill piracy and turn the comuputer into a locked down device like a console see here:
https://youtu.be/U7VwtOrwceo?t... [youtu.be]
Secure boot wasn't about "protecting your PC" from malware it was about putting hardcore antipiracy tech in your PC and turning your pc into a locked down device like the iphone and console so you n
Re: (Score:3)
I have Debian/KDE on my Pine Phone and Lineage on my Pixel.
I can see linux boot on the Pixel and get a userland easily.
What is stopping anybody from calling systemd as init and bringing up Debian on a Pixel?
The drivers are all there in the Lineage build.
So build it.
unclear on the use case (Score:2)
I've read the links. It is out of my realm of programming, but I gather that it is some kind of back end infrastructure thing that defines a cut-out of computation which will make it easier for some guys to do some things. I don't quite get what those things are.
"reduce latency and improve performance for applications that require real-time data processing", sounds good but some examples would be nice. Does this thing live in a cloud service? If so it doesn't sound very real-time.
"The proof of the power and
Re:unclear on the use case (Score:4, Interesting)
The reason the use case is unclear is because it is generally not that compelling.
A DPU is basically just another computer that happens to live on the network card. Nothing more, nothing less. It has the power to create and remove 'network ports' and either let the ports connect to outside world, or talk to it like it's a router (dedicated to only that one system).
For example, if you want to, this can allow the 'virtual switch' to run on the general purpose processor on the network card. In theory, the card could even do some hardware things tailored to handling ethernet switching, but in practice on the scale of a single network adapter, that's not really needed. So this is the 'offload' piece. However, the reality is that this task isn't *that* demanding even for high speed networks. Processing power has more than kept up with single port network throughput.
Another use case is "oh, but now you can implement arbitrary network logic in the NIC, where the NIC is a router/firewall/whatever". However the problem here is that you either can use the host CPU or the network infrastructure like you always have. Yet another processor in the middle seems a bit superfluous.
The other main point they bring in is 'security'. In other words if a guest 'escapes' virtualization, then they still can't reconfigure the networking. Which is silly because for one, that is solving a problem that hasn't been a thing, and also not really solving it because escaping the VM would mean you have access to all the virtualized ports and you could hop over to whatever network segment you wanted. Also redundant with network switch, which *also* filter traffic from outside the reach of compute. The premise is that it is both detached but also visible to the number of guests, but in practice that detachment is dubious.
In short, it's a solution looking for a problem. It's trying to carve out a new role for open ended compute when we are already flush with open ended compute in switches and traditional compute.
The downsides are that they have to be careful about power consumption, because PCIe add-on cards aren't supposed to use more than a small amount of power, so they are constrained. Operationally, you basically just doubled (or more) your overhead for managing operating systems. As it stands, DPUs represent a bit of lock in if you take them too seriously, the OS build you manage for a Bluefield, well, requires you to keep buying Bluefield family network adapters, rather than hop over to another.
Re: (Score:1)
30 years ago, NSA building computers like this (Score:2)
project (Score:1)