USENIX ATC '21/OSDI '21 Joint Keynote Address-It's Time for Operating Systems to Rediscover Hardware

USENIX ATC '21/OSDI '21 Joint Keynote Address-It's Time for Operating Systems to Rediscover Hardware

Timothy Roscoe, ETH Zurich

A glance at this year's OSDI program shows that Operating Systems are a small niche topic for this conference, not even meriting their own full session. This is unfortunate because good OS design has always been driven by the underlying hardware, and right now that hardware is almost unrecognizable from ten years ago, let alone from the 1960s when Unix was written. This change is receiving considerable attention in the architecture and security communities, for example, but in contrast, so-called OS researchers are mostly in denial. Even the little publishable OS work that is not based on Linux still assumes the same simplistic hardware model (essentially a multiprocessor VAX) that bears little resemblance to modern reality. In this talk, I'll speculate on how we came to this unfortunate state of affairs, and what might be done to fix it. In particular, I'll argue for re-engaging with what computer hardware really is today and give two suggestions (among many) about how the OS research community can usefully do this, and exploit what is actually a tremendous opportunity.

Timothy Roscoe is a Full Professor in the Systems Group of the Computer Science Department at ETH Zurich, where he works on operating systems, networks, and distributed systems, and is currently head of department.

Mothy received a PhD in 1995 from the Computer Laboratory of the University of Cambridge, where he was a principal designer and builder of the Nemesis OS. After three years working on web-based collaboration systems at a startup in North Carolina, he joined Sprint's Advanced Technology Lab in Burlingame, California, in 1998, working on cloud computing and network monitoring. He joined Intel Research at Berkeley in April 2002 as a principal architect of PlanetLab, an open, shared platform for developing and deploying planetary-scale services. Mothy joined the Computer Science Department ETH Zurich in January 2007 and was named Fellow of the ACM in 2013 for contributions to operating systems and networking research.

His work has included the Barrelfish multikernel research OS, as well as work on distributed stream processors, and using formal specifications to describe the hardware/software interfaces of modern computer systems. Mothy's current research centers on Enzian, a powerful hybrid CPU/FPGA machine designed for research into systems software.

[aiyengar]

Excellent talk. When I first start working on SoC's ~10yrs ago, I was amazed at this very revelation. The architecture has been specifically designed to isolate the "user OS" -- linux or windows on to one CPU, meanwhile there are bespoke execution environments running on a dozen different cores managing the whole H/W that traditional operating system textbooks claim to be their domain. This is such a rich space to explore and I hope to see some interesting ideas coming out of academia!

[joshmcdonald]

I don't see how this might be changed. Most of those little red circles are not just "tiny computers" but also different implementers in different companies, all trying to protect their little fiefdom against their customers, their complement components, and most especially, the users. They do this with undocumented or partially-documented interfaces, encryption schemes for which neither their customers nor users can get keys, and god knows what-else. Only somebody like Apple who can afford to own the processes and teams behind every gate in the device can even hope to address these problems. And there's not many companies with that kind of resources, and none of them are allowing users or outside-of-house developers into the machines they ostensibly "own". How might researchers (let alone hobbyists) even *dream* of being able to do novel work in such an environment?

[ewetoo]

The same spirit that fueled Linux can also fuel a proper SoC OS whatever form that takes, but Linux is now a roadblock, or rather its mindset has become one. I hope someone sees this and takes that spirit and creates something awesome with it.

[DrewryPope]

Tim woke up and chose violence. lol


no but really this is great, a bit opinionated but many good things are. here's hoping we're looking at an open future in embedded systems

[harisha.g]

A much needed talk indeed ! It was very good...Thanks :)

[mondovicium]

NetBSD at least attempted to address management of interfaces between hardware components with its Bus Space abstraction. I'm not sure how successful it was, but at least some of the problems have been known for a while.

[mysticknight]

Part of the formal spec of the OS needs to be the actual Verilog or VHDL specification for the SOC — however, this needs to be extended to CXL at a data center scope with heterogenous arrays of processors

[mamborambo]

In my University days they were talking about microkernels, message passing kernels, transputers, etc. Seems like OS development has been stagnant for too long.

[shenidan]

Excellent and thought provoking presentation

[jasonsmethers]

Some distributed systems applies the required rigor with designs specified in state provers. You also see this on the EE side with similar provers using models, FPGA simulation, and execution for chip design. This is unlikely to become common practice because it requires too much expertise and effort to do. The hardware guys are trying to sell hardware which might be better done as software and the software is a complete afterthought to this. Even the hardware design by the EE's often ignores how the software will or should interface with it. Plus everyone is frozen with supporting and extending whatever standards and cruft that already exist. There seems to have definitely been an improvement in the last couple of decades with improvements in software engineering practices and application of compute science theory and principles, but it still seems like you'll only see a select few founding people working on platforms like this and then it slowly degrading as other people join in on their projects. Why are basic things like parts of POSIX not depreciated and superseded? You cannot get people to stop using this junk and replace it. Maybe this will improve with hypervisors pushing this kind of separation? Who is going to pay for these improvements? Who is going to take credit for doing them? Are you sure everyone isn't just a Java developer with limited knowledge of what a byte code VM with preallocated memory might be doing?

[argxuy]

OS is a big topic. Like some companies bring new hardwares, which can make a difference to OS. So those papers that deal with the impact of new hardware on OS should also be OS papers.

[mysticknight]

Very few commenters getting the point — (1) how to make a single security policy framework work across multiple components that speak different languages, (2) same for resource management and methods for different components to share resources without stomping on each other … and all beginning form a zero-trust context

[xerzy]

You should be more around fedi (Mastodon and company). There's plenty of demand, as well as people with motivation, but the latter just either lack the technical background or (most likely) the out-of-the-box thinking. And they're all looking to have fun! So you have to teach, like in this video, which assumptions exist and why ‒ but particularly, how software simplicity is still possible, instead of giving up and refusing to acknowledge these kinds of commonplace SoC/SiP designs because they're more complex and thus not as fun or seemingly accessible.

[MC_CatMaster]

I think he's onto something, when can I have Spongebob Krusty Cookoff as my OS?

[blacklion]

I've seen first 5 minutes only and it is very depressing.
Linux is not great OS, but its good-enough-ness kills whole field.
I start to hate Linux in last years not because I'm MS or Apple fan, but because I think, that Linux monoculture is very bad and dangerous.
Linux for storage servers, Linux for databases, Linux for routers, Linux for places where we need true realtime, Linux for smart bulbs, Linux everywhere, where specialized systems will be better.
I really sad about other UNIXes (Solaris was much better designed, BSD family is much nicer for novice, in my opinion), but no, it's only Linux.
It is not only about UNIXes, it is about all other OSes.
It likes there are no any mechanic's instruments anymore but multitool from Leatherman. Yes, you could fo almost anything with multitool, but nothing with comfort! Linux looks the same.

[fredconcklin]

The death's tar would Blockchain and usenix

[darelcullen]

I wouldn't blame linux or insert-your-os for this, this is down to the OEM hardware manufacturers, they design the hardware with heterogenous cores. This is just a failure of Hardware and Operating system designers collaboration.

[Gooberpatrol]

Notice how he keeps saying the OS "doesn't do what hardware manufacturers need" but never actually elaborates on what it is that is impossible to implement in the kernel. There is one reason only SOC manufacturers try to hide from the OS instead of making it visible and writing a linux driver for it: protecting their trade secrets. The "missing feature" is the ability to hide from the users and customers. This man is suggesting (well, wishing for; he proposes no actual solutions himself) an engineering solution to an economic problem.

[blacklion]

Does HarmonyOS share code with Linux for real? Looks like it is still Android-based.

[jameselsing]

Linux and Unix still work mainly with binary... same are PCs