Jon Masters is in an odd position—he is the chief ARM architect at Red Hat. Since Red Hat must make almost all of its money from selling operating systems and services for x86-based servers, this is definitely a minority interest position. However, he went out of his way to say that he wasn't there to sell Red Hat to us. In fact, he was in even more of a minority interest position this time. For the last couple of years, ARM TechCon was all about servers and the hope for penetration of ARM-based servers into the explosively growing cloud market. The technical numbers look good: a tenth of the cost, a tenth of the power, a tenth of the physical size. Apparently there is more stuff moving in China than the US, but still, ARM's share of datacenters must be about the same as...well... Intel/Microsoft's share in mobile. This year the focus is very much on the internet of things (IoT). In fact, you would hardly even notice that ARM is dominant in mobile, never mind that they hope to get a foothold, at least, in datacenters. I have seen Jon speak before, at a Linley Datacenter conference. See ARM Servers Can Take Over the World . There he gave a timeline of Red Hat's involvement in the ARM server market that I think is worth repeating. 2011: Red Hat ARM team formed, industry standardization effort begins, secret RED Hat ARM v8 OS bootstrap begins, ARMv8 architecture announced, Red Hat on stage with AppliedMicro (showing X-Gene) 2012: Many design collaborations initiated, Linaro Enterprise Group (LEG) started, OpenJDK initial release 2013: ARMv8 hardware arrives at Red Hat, world's first public demonstration, Broadcom announces Vulcan ARMv8 server processor 2014: ARM server-based system architecture (SBSA), ARM server-based boot requirements (SBBR), Red Hat on stage with Cavium (ThunderX), Red Hat demonstrates rack-level provisioning and launches ARM early access program 2015: Ceph Cluster (AppliedMicro X-Gene, AMD Seattle, Cavium ThunderX, and others), Red Hat Enterprise Linux Sever 7.1 and 7.2 development previews, Qualcomm announces 24-core prototype server SoC That's a lot of activity to which one can only add "2016...not much." In fact, Jon was trying so hard not to sell us on Red Hat that he barely mentioned it. His focus was on The Wonderful World of Open Source: How Collaborative, Community-Powered Innovation Is Changing our Modern World . He had the numbers to prove it. there are 10M developers working on open source. There are 26 million projects on Github. He got us all to guess the #1 contributor to GitHub today and we were all surprised that it was that stronghold of closed-source software, Microsoft. I believe that open source is changing the world, but only part of it. To me, it seems that open source is mostly about programmers "scratching their own itches" as Jon put it. The best tools for software development are all open source. It is not clear if that is true in other areas. I once saw Eric Raymond (esr) speak and he pointed out that there are no open source games. By the time that it is obvious that a game is a hit, it is too late to clone it as an open source project since gamers will have moved on by the time there is a deliverable product. So there is lots of open source gaming infrastructure, such as physics engines (arguably, the heart of Angry Birds is the physics engine, but Rovio didn't write that, Erin Catto did, and it was open source). EDA may fall into the same category. There are very few open source EDA projects, and where there are projects, they tend to be in the most static areas of EDA such as RTL simulation. Also, open source EDA projects face one of the big issues that all EDA startups face: test data. It is easy to get access to out-of-date designs, but if you want to sell to the leading edge, then you need leading-edge designs. This is a challenge for two reasons. First, the big guys are not going to give you any data unless you have something amazingly compelling, and you probably don't. Plus there is the challenge that even a company like Cadence faces. We have to develop tools for 7nm (and even 5nm), but there are no test cases because nobody has done any designs yet, at least partially be cause there are no tools...rinse and repeat. What we call open source today was really started by Richard Stallman (rjs) and the Free Software Foundation. He would always emphasize that it was "free as in freedom, not free as in beer", but since it was usually free as in beer, too, that message got a little confused. Eric Raymond invented the term "open source", which seemed to make things more acceptable in the industrial environment. Nobody really knew just how far open source would go. I don't think anyone predicted that Linux would power almost all datacenters. Linus Torvalds certainly didn't. His initial email said "I'm doing a free operating system, just a hobby, won't be big." That was 25 years ago. One reason that open source is so successful is Linus's Law: "Given enough eyeballs, all bugs are shallow." This was actually formulated by Eric Raymond and named in honor of Linus. What it means is that with enough people looking at the code, any complex problem will be obvious to someone. That "looking at the code" part is important, and requires the code to be available. Not matter how aggravating a bug you run into in, say, Microsoft Word (or Genus Synthesis Solution, for that matter), you don't get to find it or fix it since the code is not available. Jon talked about how he got involved in open source. He recalled Eben Upton's talk the day before about Raspberry Pi (see my post The Amazing Raspberry Pi Story ) and he also remembered having a BBC Micro in his bedroom. Actually, he got access to a lot of kit a lot earlier than most of us, since he went to University aged 14 to study Computer Science. He wanted a Sun workstation when he was a 14 year old with no money. Someone gave him a copy of the original version of Linux, Slackware, in 1996. It took him downloading 200 floppy disks (several times) to get it installed. Now, at Red Hat, bringing up Linux on a virgin server is five minutes work. He ran into some problem so he turned to the Oxford Linux Users' Group (OxLug). He was invited along, and everyone would learn a lot. And then everyone would go off to the bar...but he couldn't since he was only 14. He started playing around with embedded Linux. At one point he got a random email about porting, since he was porting Linux to FPGA-based processors. He ended up with a book contract (I assume Professional Linux Programming , which is obviously now a collector's item since it will cost you over $200 to acquire a second-hand copy). As a result of the book he got an offer to work at Red Hat. He started playing with ARM and realized it could be big. Obviously this was in collaboration with many people, such as the Linaro community pictured above. He is convinced it is going to be big. After his presentation, he went to the pavilion in the exhibit hall and took questions. The first question was about open source hardware. He pointed out that hardware, especially chips, has very large numbers associated with it. With software, you can mostly fix any problems later (that may be true for some software, but it is certainly not how, for example, automotive software is developed). Of course, there are lots of open source projects that use Arduino and Raspberry Pi. A follow-up question was about modems and Jon pointed out that there are compliance issues. He didn't talk about it, but now it is going to be illegal to have open source code for your WiFi router since if you get changes wrong you can take down all the wireless in the area. Jon told a story about when he was once asked about cars. He pointed out that once when he was with Richard Stallman, he asked whether Richard though he should be able to change the code on his microwave oven. Richard thought that was crazy, but probably mostly because he is not interested in embedded software. Jon asked the person asking about cars who said "I believe I should get the source code and be able to make any modifications I want." Even Jon had his limits on the desirability of open source, and said he would not support that. Partially because of the danger. But also, if a Ford (for example) crashes, the news will be "Ford crashed", not guy "modified the software and screwed up." Jon was asked about servers, and a remark that Linus made that apparently he prefers x86. He said that was an old remark from when things just ran on x86 and required a lot of tweaking for any other architecture. Jon has put a lot of effort into making servers "boring." People used to show up saying "here's my special version of Linux" and he'd say "here's my special version of 'go away'." He acknowledged that it has not been easy: “Those of us pushing the ARM boulder up the hill for five years, to get it over the top, know it is a lot of work.” Previous: Optimizing Power with Palladium![]()
![]()
↧