Biz & IT —

Linus Torvalds worries about how Linux will handle end of Moore’s Law

Top kernel devs say mobile devices are driving Linux's present and future.

Linux kernel developer panel with (from left to right) Sarah Sharp, Tejun Heo, Linus Torvalds, and Greg Kroah-Hartman.
Linux kernel developer panel with (from left to right) Sarah Sharp, Tejun Heo, Linus Torvalds, and Greg Kroah-Hartman.

Perhaps the most hotly anticipated event at LinuxCon was this morning's kernel developer panel featuring none other than Linux creator Linus Torvalds.

In the panel, Torvalds said he's worried the possible end of Moore's Law might finally be within sight, providing challenges to both hardware and software developers.

"On the five- to 10-year timeframe scale, I'm very interested to see how the industry actually reacts to the fact that soon we will come against some physical limits," Torvalds said. "People used to be talking about having thousands of cores on one die because it keeps shrinking, and those people clearly have no idea about physics because we won't be shrinking for much longer."

Both physical and financial limits could prevent the frequent doubling in transistor density that was observed by Moore's Law, he said. The impact of hardware advances on Linux software has been huge. "The reason Linux runs really well on cell phones is cell phones grew up" and are now thousands of times more powerful than the first machine Linux ran on more than two decades ago, Torvalds said.

"In five, 10 years it's going to be tough," he continued. "That's going to affect us in kernel land because we are the layer between hardware and software. What happens when hardware doesn't improve and magically make us faster? That's going to be interesting. It might not be five or 10 years, it might be 15, but it's going to happen."

Torvalds hopes hardware manufacturers can keep improving their side of the equation even if increases in processing power slow down. "Hopefully hardware innovation doesn't stop just because shrinking stops," he said.

From mobile to the data center, Linux is everywhere

Torvalds was joined by Linux kernel maintainer Greg Kroah-Hartman; Intel developer Sarah Sharp, who works on Linux's USB subsystem and first got involved in Linux years ago when she built USB sensors for a Linux-based rocket; and Red Hat engineer Tejun Heo, who maintains several parts of the kernel. Ric Wheeler of Red Hat was the moderator, passing the audience's questions on to the panelists.

Linux has to keep changing to adapt to different form factors, Kroah-Hartman said. "My big goal is that I want to see Linux continue to succeed. We have to continue to change to work with all the hardware and devices out there," he said. "If we stop changing, we will die."

Much hardware innovation is happening on devices that aren't necessarily amazing under the hood, Sharp said. For example, "the hardware [in Google Glass] isn't that advanced, it's basically a PandaBoard but what you do with it is very interesting," she said. "It's going to be interesting to see what we end up doing with Linux and where it goes, and what we end up doing in userspace with it."

The panelists described how embedded systems, particularly mobile devices, are driving Linux development. But even advances in mobile can benefit the data center, they said.

"Embedded today is what enterprise was five years ago," Kroah-Hartman said. "You have a quad-core in your pocket. The fun thing about Linux is all the changes you make have to work on all the things."

The advances in power management driven by mobile devices initially weren't that interesting to enterprise developers, according to Kroah-Hartman. That quickly changed once they realized it was helping them save millions of dollars in data center power costs.

In the last merge window, most of the code was on the embedded side because of all the drivers for various devices, Torvalds said. But there was also a new feature for "scalability to 200-core machines that people are looking at in the corporate world," Torvalds said. "We have a pretty good balance, I think."

Sharp said the advances in embedded mean she's more frequently testing code on devices beyond the desktop. Kernel maintainers have to ask, "Is it going to break the embedded side?" when evaluating code, she said. "It's definitely more prominent in my mind these days."

While the Linux kernel is low-level software that many users might not even be aware of, the panelists touched briefly on the interactions between the kernel and "userland" or "userspace." In particular, Sharp said kernel developers should collaborate more with userspace developers on power management. Even if the kernel has the greatest power management in the world, poorly written programs in user-focused applications can drain battery, she noted.

Torvalds complained that many userspace projects break compatibility with other software, which is something he's steadfastly avoided in the kernel itself. "I'd really like to see some of the kernel culture spread into userspace," even if that means some developers leave the kernel project to go work in userland, he said.

Collaborative cursing

Linux development happens all out in the open, putting disputes and the various idiosyncrasies of contributors into the public eye.

Torvalds has long been known for the creative insults, threats of violence, and curse words he hurls at developers who give him what he deems to be bad code. Sharp recently called him out, saying, "Keep it professional on the mailing lists."

Torvalds replied that he is "not interested" in acting professional and attributed his frequent cursing to his Finnish culture. "I simply don't believe in being polite or politically correct," he wrote.

Torvalds hasn't changed course. Earlier this month, he expressed hope that some ARM SoC hardware designers "all die in some incredibly painful accident."

“So if you see any, send them my love and possibly puncture the brake-lines on their car and put a little surprise in their coffee, OK?" he wrote.

Such conflicts were never mentioned in today's panel, but Sharp got a round of applause when she said she hopes for a more welcoming environment to newcomers. "I'd like to make sure our community is inclusive to all people that want to contribute," she said. "Getting more diversity, getting more diverse voices in our community is something I'd like to see."

Earlier in the discussion, panel members talked about some programs designed to mentor new developers, including one program specifically for women. They'd like new developers to become frequent contributors instead of just contributing one patch before stopping.

Although kernel development can be intimidating because of how big and complicated the project is, Kroah-Hartman noted that recently a professor in the Czech Republic required a classroom of students to get patches merged into the kernel. A few of them liked the process so much they said they'd keep contributing, he said.

Torvalds said he's happy with the level of involvement. "The kernel has a ton of developers and we get patches from 1,000 people every single release, while many other open source projects are struggling to find five people who are involved," he said. "People talk about how hard the kernel is but just look at the numbers. It can't be that hard to get involved."

New developers need to read the provided instructions and make sure they contribute in a useful way though. Sharp said her pet peeve is people contributing patches without explaining why they are necessary.

Torvalds agreed, adding that developers should avoid submitting last-minute changes right before a release, and that they need to make themselves available to answer questions. "Another thing I personally find hard to deal with are these drive-by shootings when someone throws a patch over the fence and doesn't understand that in order to be a useful kernel developer, it's not enough to write the code and send it out there," he said.

A couple of final notes: The panelists were asked if they've been approached by the US government to insert a back door into the Linux kernel. Torvalds said no while nodding his head, drawing a laugh.

The panelists were also asked if they're interested in the open CEO job at Microsoft. None of them were.

Channel Ars Technica