What happened this year?
First of all, Linaro is 7. I find it hard to believe as it has been quite a journey and (as usual) has consumed all of my work energies. From a single issue organisation (the Linux kernel code base for the Arm architecture is a mess) Linaro has evolved into an organisation of over 30 members contributing to almost all areas of the Arm ecosystem’s activities. More on ‘almost’ later.
2017, for me at least, has felt like a transitional year for the Arm ecosystem, including Linaro. One transition is that it was traditionally rooted in mobile; this year has seen a shift towards enterprise and the cloud, and great progress has been made in LEG towards Arm servers being deployed. The real value that the Arm ecosystem brings is that it generates rapid Darwinian evolution; trying many solutions simultaneously. Often the key is hardware acceleration, although this must be integrated with the right software to deploy. You can see this strongly in enterprise computing, but also in networking, mobile - well, everywhere really.
Another transition (although of attitude and not technology) has been the Softbank acquisition of Arm. This has not altered Arm’s IP and architectural licensing business (where it is business as usual as they continue to invent world beating technology), but positive effects are starting to be seen in their activities in other areas, especially around the cloud and internet of things. Meeting with Softbank (one of the reasons for several trips to Tokyo this year) has been really interesting.
Linaro’s segment groups can be confusing, so rather than enumerating segment groups I prefer to think about 3 areas of activity. That is, client, fog and data center. Client computing includes set top boxes (LHG), mobile (LMG) and embedded systems (LITE). Fog networking is potentially owned by LNG. Data center includes LEG and High Performance Computing (HPC). Technologies such as machine learning fit across that spectrum as do technologies such as boot architecture, security and power management.
Mobile client computing is maturing and you can see that in LMG, with most companies concentrating on refining the flow of technology into products. Linaro’s Sharp project (which supports Google’s project Treble) is all about the supporting the Android AOSP. That is not to say that the power management work is not important, we need our mobile devices to last all day. Mobile devices are where most of us are seeing machine learning as personal assistants get more personal. LITE’s work on the Zephyr project continues to show progress both in establishing a maintainership and in code functionality and stability. Most of the SoC vendors that I talk to have plans to include Zephyr, mainly for use in IoT. In my mind, the key thing about Zephyr is not the code, it is the maintainership. Establish the right maintainership and you will get the right code and we’re seeing this in Zephyr.
Linux and IoT deserves some attention. After all, Linux is a reservoir of code and experience that can be leveraged for IoT clients. Large parts of Nicolas Pitre’s work on IoTL (IoT for Linux) has been merged upstream. A real achievement, especially as Linux development is largely rooted in big iron. This shows how Linux can be super-slimmed down for practical use within very constrained IoT clients (in this case a Cortex-M based device). Watch out for Nico’s LWN articles on IoTL shortly.
Through the year I have seen a number of RTOS acquisitions as both cloud ecosystem providers and silicon creators seek to provide a floor to ceiling solution for integrating devices with the cloud. Thinking that you can control things via a proprietary client operating system is an illusion. The real need is to allow as many devices (hopefully Arm based) to be connected to as many cloud ecosystems as possible and I don’t see that happening without publishing freely reusable code for clients, gateways (fog) and cloud.
The Arm ecosystem is very much in a ‘get to market’ mode for the data center and great strides are being taken. Good to see public announcements of real products with production operating system support. Part of the Arm ecosystem’s journey into the data center has been to move to a standard boot architecture. A different standard to mobile devices, but a standard nonetheless. I have previously predicted that technologies will migrate from the data center to networking, but whilst I still think that is true, it will take a while for the largely embedded community to move away from their trusted and known boot agents to UEFI, for example. The move to containers continues apace as they are an essential ingredient for transactional updates to systems, itself part of an overall security strategy. OpenStack is a good example of containerisation. An interesting side effect of containerisation is that it allows the producers of technology to directly ship that technology and, to an extent, undermines the traditional role of operating systems. On the other hand, operating systems are really just technology ingredients and the companies that make use of them use them in order to provide a set of services and guarantees. I have just got back from Tokyo where I attended the Arm HPC Workshop by RIKEN AICS and Linaro last week - the Arm ecosystem will bring a lot of value and change to HPC. HPC is, for me, a bit of a ‘back to the future’ area. It really is about big (really big) iron and bespoke systems. These bespoke systems are running Linux though, and the key compatibility is the code base.
Fog networking is ‘the bit in the middle’ and, in my view, the most interesting area of all, as it is very important with the biggest potential for change. Think of it this way, we are building a cloud based ecosystem supporting mobile applications. This covers the Android application on your mobile, an example of which is a Nest thermostat.. This ties your house’s thermostat to Nest’s cloud to the Nest Android application. Relying on the cloud to process data (analysing your home’s heating patterns against use and local weather, for example) and manage your heating will not scale to the level of automation that we need outside the home. We cannot expect factories to ‘ask the cloud’ every time they want to heat the beer mash in a brewery. We need local processing and local storage. This is where fog networking (and a bunch of complexity) comes to the rescue including the thorny problem of how computing and storage agents find and trust each other (think blockchain).
Fog networking will transform many areas of technology, not least traditional edge networking. There continues to be huge fragmentation in traditional edge networking but the need to standardise secure over the air (OTA) updates will drive standardisation in this area. In Linaro we have done some work around a standard gateway platform. This is based on Yocto / Open Embedded and uses a thin layer to provide deployment services - ansible is one example and container support is another. We hope that EdgeX Foundry will help standardise how gateways are deployed and managed. Other aspects (the payload as it were) will be driven by the segment groups / markets.
I include the work that LHG has done integrating media decode in tandem with open trusted execution environment (OP-TEE) as part of standardising the gateway. Again Yocto / Open Embedded is the basis for that work.
LNG has focused on OpenDataPlane (ODP), but in 2017 it has widened its scope to include real time Linux extensions. It’s scope will also include fog networking as that evolves. A key part of LNG’s work is to understand how the Arm ecosystem can support advanced secure high speed networking whilst at the same time allowing the use of hardware acceleration. Key questions are which APIs do you develop and support to best deploy hardware acceleration.
Back to ‘almost’. I spent a lot of time in 2017 looking at automotive and one clear takeaway is that Linaro can help here. Technically (and it’s obvious when you think about it) a car is a robot on wheels. Less obviously, it is where machine learning will be first applied directly to our lives in a safety critical way. By contrast, whilst I like Google Home and Alexa, generally they cannot harm me directly, whereas my car (or someone else’s) can. It is not just automotive though, drones and robots have a lot in common. Automotive is being incubated by Linaro’s Technical Steering Group.
Whatever you do and wherever you are (and especially if you are a member of Linaro), have a good holiday and a great 2018.