WEBVTT 00:00:00.001 --> 00:00:08.020 When you think about the performance of your software, there's nothing more low-level and fundamental than how your code executes on the CPU itself. 00:00:08.020 --> 00:00:16.820 Many of us study and try to understand how to maximize performance at this low level, but few are in a position to define what actually happens there. 00:00:16.820 --> 00:00:27.740 That's why I'm thrilled to share the work that Intel, the largest PC chip manufacturer, is doing specifically to make Python faster and to make their chips execute Python code even better. 00:00:28.240 --> 00:00:34.360 This week, you'll meet David Stewart, Engineering Manager in the Intel Data Center Software Technology Group at Intel. 00:00:34.360 --> 00:00:38.980 We'll discuss a wide variety of work Intel is doing in open source and Python. 00:00:38.980 --> 00:00:44.900 This is Talk Python To Me, Episode 57, recorded May 2, 2016. 00:00:56.480 --> 00:01:01.660 This music I constructed line by line, just like when I'm coding another software design. 00:01:01.660 --> 00:01:04.860 In both cases, it's about design patterns. 00:01:04.860 --> 00:01:06.360 Anyone can get the job done. 00:01:06.360 --> 00:01:07.880 It's the execution that matters. 00:01:07.880 --> 00:01:09.360 I have many interests. 00:01:09.360 --> 00:01:16.160 Welcome to Talk Python To Me, a weekly podcast on Python, the language, the libraries, the ecosystem, and the personalities. 00:01:16.160 --> 00:01:18.280 This is your host, Michael Kennedy. 00:01:18.540 --> 00:01:20.280 Follow me on Twitter, where I'm @mkennedy. 00:01:20.280 --> 00:01:24.180 Keep up with the show and listen to past episodes at talkpython.fm. 00:01:24.180 --> 00:01:26.760 And follow the show on Twitter via at Talk Python. 00:01:26.760 --> 00:01:31.720 This episode has been brought to you by SnapCI and now Metis. 00:01:31.720 --> 00:01:37.220 Thank them for supporting the show on Twitter via Snap underscore CI and this is Metis. 00:01:37.220 --> 00:01:38.000 That's right. 00:01:38.000 --> 00:01:42.220 Metis and their data science education team have joined the show to keep the episodes coming. 00:01:42.720 --> 00:01:44.700 Be sure to find them on Twitter and tell them thank you. 00:01:44.700 --> 00:01:47.260 David, welcome to the show. 00:01:47.260 --> 00:01:48.940 Thank you very much, Michael. 00:01:48.940 --> 00:01:50.140 It's great to talk with you. 00:01:50.140 --> 00:01:53.540 Yeah, I'm really excited to have you on the show today. 00:01:53.540 --> 00:01:59.040 And I'm looking forward to looking inside what you guys are doing at Intel with Python. 00:01:59.040 --> 00:02:00.160 It's going to be a lot of fun. 00:02:00.600 --> 00:02:03.180 Yeah, it's really been exciting to me. 00:02:03.180 --> 00:02:12.160 I think people have been really responding when I hear that Intel is doubling down on Python just because of its power and popularity. 00:02:12.160 --> 00:02:15.940 So I think it's exciting to be a part of that for sure. 00:02:16.520 --> 00:02:19.320 Yeah, it's really exciting to be part of Python in general. 00:02:19.320 --> 00:02:30.740 I mean, it's really surprising and a pleasant way to me to continue to see this language and ecosystem that's 25 years old gaining momentum and gaining speed. 00:02:30.740 --> 00:02:35.000 Kind of feel like it would have done whatever it did in the first five or ten years, but that's not the case. 00:02:35.000 --> 00:02:35.700 It's great. 00:02:35.880 --> 00:02:36.900 So nice to see Intel. 00:02:36.900 --> 00:02:48.800 Your listeners probably, you're probably already aware of the fact that the top, the majority of the top CF schools in the U.S. are teaching Python as the first introductory language, right? 00:02:48.800 --> 00:02:55.600 That's often a surprise when I talk, I was just talking to a professor at computer science last week and he was like, really? 00:02:55.600 --> 00:02:57.100 They just didn't know about that. 00:02:57.100 --> 00:02:58.820 So, yeah, it's really taken hold. 00:02:58.820 --> 00:03:00.180 Yeah, I think that's great. 00:03:00.180 --> 00:03:02.540 And, you know, to be honest, it makes me a little bit jealous. 00:03:02.540 --> 00:03:12.760 Yes, I, my very first CS, not my first programming class, but my first CS class that was for programmers was in Scheme and Lisp. 00:03:12.760 --> 00:03:14.540 And I would have definitely preferred. 00:03:14.540 --> 00:03:14.740 Oh, my gosh. 00:03:14.740 --> 00:03:17.420 Yeah, I would have definitely preferred to get Python in there. 00:03:17.420 --> 00:03:18.720 But, you know, you get what you get. 00:03:18.720 --> 00:03:19.320 Yeah, no. 00:03:19.320 --> 00:03:21.260 I know what you're talking about. 00:03:21.260 --> 00:03:25.220 Well, you know, if you were talking about Pity here, my first was with Fortran. 00:03:25.360 --> 00:03:33.800 And, in fact, I had to teach Fortran programming to, you know, to engineering students and the like as a part of a graduate teaching assistantship back in the day. 00:03:33.800 --> 00:03:42.140 So, yeah, I didn't necessarily do a lot of programming in Fortran myself, but it was one of those things where you sort of had to cut your teeth on something. 00:03:42.140 --> 00:03:46.400 And there's still people using Fortran these days, but it's probably more than Lisp. 00:03:46.400 --> 00:03:48.900 But, yeah, it's got to be waning. 00:03:49.360 --> 00:03:50.100 I hope so. 00:03:50.100 --> 00:03:52.480 I think we share some common history. 00:03:52.480 --> 00:03:59.520 Like, I started out in, yeah, I started out in an engineering field, not in computer science. 00:03:59.520 --> 00:04:03.000 And when I got there, they said, you have to have a programming class. 00:04:03.000 --> 00:04:04.100 I said, great, can I take C++? 00:04:04.100 --> 00:04:06.120 No, you have to take Fortran first. 00:04:06.120 --> 00:04:09.380 This is the most important language you will ever learn in your life. 00:04:09.380 --> 00:04:11.780 Then you can go take those other less meaningful ones. 00:04:11.780 --> 00:04:15.300 So I took that and then I went to CS. 00:04:15.300 --> 00:04:16.820 I said, can I please take C++? 00:04:16.820 --> 00:04:18.400 No, you have to take Scheme. 00:04:18.400 --> 00:04:20.740 I'm like, why can't I do a real programming? 00:04:20.740 --> 00:04:24.200 Anyway, I'm going to get some hate mail for saying that, but that's okay. 00:04:24.200 --> 00:04:25.240 I'm sure you will. 00:04:25.240 --> 00:04:32.400 No, I really, I do like C++, but I wish I had Python in the beginning. 00:04:32.400 --> 00:04:36.420 That would have inspired me more, even though I was already inspired to do programming. 00:04:36.420 --> 00:04:40.840 Yeah, and I think that it's been particularly interesting. 00:04:40.840 --> 00:04:47.760 There's so many open source projects that are, you know, kind of advanced by some corporate entity or something like that. 00:04:47.760 --> 00:04:50.340 Python is a phenomenal project. 00:04:50.340 --> 00:04:53.100 It's sort of like the Linux kernel itself, right? 00:04:53.100 --> 00:05:01.860 There are plenty of companies involved with Linux, but it's really, you know, with Linus Torvalds being the, you know, the lead, 00:05:01.860 --> 00:05:03.320 it makes a final decision on things. 00:05:03.320 --> 00:05:09.100 And Python, that's one of the strengths, I think, is the kind of approach that Guido uses. 00:05:09.220 --> 00:05:11.980 Guido Van Ropsen, who is the benevolent dictator for Life for Python. 00:05:11.980 --> 00:05:20.880 I think he's really created, you know, just kind of an ethos and a culture around this project that I think is super unique. 00:05:20.880 --> 00:05:29.060 And, in fact, it's a great, I hold up as a great example for people who aspire to, you know, establishing open source projects. 00:05:29.180 --> 00:05:34.480 I mean, hey, this is one of the ways that you talk about, yeah, 25 years of, you know, experience. 00:05:34.480 --> 00:05:38.240 And Guido's, you know, I think a phenomenal leader for that movement. 00:05:38.240 --> 00:05:39.020 Yeah, for sure. 00:05:39.020 --> 00:05:40.380 Yeah, he's doing a great job. 00:05:40.460 --> 00:05:47.580 And I think there's a lot of positive examples of the Python community being a very positive place for open source. 00:05:47.580 --> 00:05:48.520 And that's great. 00:05:48.520 --> 00:05:55.360 So, before we get into the story about what you guys are doing with Python, maybe you could just sort of tell me how you got into programming in Python. 00:05:55.360 --> 00:05:58.220 You said you took Fortran in college. 00:05:58.220 --> 00:05:59.360 So, how did you get to there? 00:05:59.360 --> 00:06:10.440 Yeah, I mean, you know, when I got my bachelor's and master's in computer science, I mean, C was really the most interesting language. 00:06:10.440 --> 00:06:12.720 to me at the time because I was doing a lot in operating systems. 00:06:12.720 --> 00:06:20.820 And so, I've been probably over the years of my career probably more interested in, you know, operating systems in general instead of languages per se. 00:06:20.820 --> 00:06:27.640 And since C is a great language, you know, initially when I worked with it was Unix and then, you know, Linux, obviously. 00:06:27.640 --> 00:06:34.380 I think that's one of the things that it was just when I think in terms of programming, it's still hard for me not to relate back to C. 00:06:34.560 --> 00:06:41.640 But, you know, the Python work came as a result of a couple of things I got involved with back in 2010. 00:06:41.640 --> 00:06:48.560 What happened was I was asked to start up a new project here at Intel to work on embedded Linux. 00:06:48.560 --> 00:06:51.780 And it was kind of an amazing thing. 00:06:51.840 --> 00:06:56.060 The core of this project had this something called BitBake. 00:06:56.060 --> 00:07:02.900 And if you're not familiar with it, what BitBake does is it's a desktop application that will build a complete Linux distribution. 00:07:02.900 --> 00:07:08.240 Now, this is actually quite interesting because what it does is it builds the compiler first, right? 00:07:08.500 --> 00:07:13.700 And it has to bootstrap several different paths of the compiler building to get a functional compiler, right? 00:07:13.700 --> 00:07:22.040 And then it goes and a Linux distribution could be made up of, like, literally a thousand different projects all over the Internet. 00:07:22.040 --> 00:07:26.740 Individual projects with their own source control systems, their own repositories. 00:07:26.740 --> 00:07:33.380 And this system will actually download the sources from all these different projects, goes and patches the source. 00:07:33.380 --> 00:07:37.820 Of course, if you have an open source system, you want to be able to, you know, do patching of those things, right? 00:07:38.120 --> 00:07:46.260 Then it will configure them, build them, build a package repository, build a running Linux image, and then an SDK. 00:07:46.260 --> 00:07:51.400 And the amazing thing is that it does all of this in an hour, like one hour on a desktop computer. 00:07:51.400 --> 00:07:53.360 And you're going, wait, hold on. 00:07:53.360 --> 00:07:56.420 What magic is this thing doing? 00:07:56.420 --> 00:08:01.520 And I started talking to the architect of this thing, and it's a massive Python program. 00:08:01.520 --> 00:08:04.860 So BitBake is actually completely written in Python. 00:08:04.860 --> 00:08:07.160 I was like, you know, I got to confess. 00:08:07.360 --> 00:08:09.780 I was like, holy cow, this is a scripting language? 00:08:09.780 --> 00:08:15.060 And I heard about, you know, things like, you know, you're able to serialize and deserialize objects. 00:08:15.060 --> 00:08:17.960 And it's like, hang on, this is really powerful stuff. 00:08:17.960 --> 00:08:21.720 So I worked on that project for about five years. 00:08:21.720 --> 00:08:23.280 That was a project called the Yocto Project. 00:08:23.280 --> 00:08:26.180 So anybody who knows that, that was something I was involved with. 00:08:26.300 --> 00:08:29.740 And then, you know, things were going, you know, pretty well. 00:08:29.740 --> 00:08:34.500 And I kind of get, you know, a little bit, you know, sedate, I guess, when things are going too well. 00:08:34.500 --> 00:08:36.020 So I always like to have a challenge. 00:08:36.020 --> 00:08:41.340 And at that point, Intel was really interested in, as I said, doubling down on Python, 00:08:41.760 --> 00:08:48.520 because it's primarily the core of a lot of key things that Intel is invested in in other areas. 00:08:48.520 --> 00:08:53.800 And so, and besides its overall popularity as a, you know, as a programming language. 00:08:53.800 --> 00:08:59.640 So that was when I really, you know, got personally very invested in what was going on in Python. 00:08:59.640 --> 00:09:04.460 So I, you know, started a group and we have engineers that are at the core of working on that. 00:09:04.460 --> 00:09:10.000 So that was where I think I got really a lot of Python religion, obviously, at that point, 00:09:10.000 --> 00:09:16.720 was not only seeing it as a user, incredibly powerful, but also be able to now, I think, 00:09:16.720 --> 00:09:19.340 affect it positively on a community basis. 00:09:19.340 --> 00:09:24.320 We're going to dig into some of the cool stuff you guys are doing on a community basis as well. 00:09:24.320 --> 00:09:32.520 But you sort of started out with the philosophy of it's in Intel's interest to sort of understand 00:09:32.520 --> 00:09:38.000 modern cloud computing and the languages that drive that and make that stuff go really well. 00:09:38.000 --> 00:09:38.780 That's correct. 00:09:38.780 --> 00:09:46.000 I mean, if you think about, you know, Intel, obviously, we try and sell processors in the data center space. 00:09:46.000 --> 00:09:47.300 I think we've had some success there. 00:09:47.300 --> 00:09:53.120 One of the things that if you look at overall, how are people programming those data center processors? 00:09:53.400 --> 00:09:58.160 Well, there's obviously a lot of C++, C#, and Java, no question. 00:09:58.160 --> 00:10:06.720 But as we kind of analyzed things, we said, well, gee, there are 7 million PHP programmers out there. 00:10:06.720 --> 00:10:12.740 There's like 9 million JavaScript programmers, a considerable number on, you know, on data center computers. 00:10:12.740 --> 00:10:14.960 And then Python is huge in that respect, too. 00:10:14.960 --> 00:10:22.340 So if you look at the top languages that people are using our processors to run, right, it sort of says, well, 00:10:22.480 --> 00:10:25.440 how do I, you know, I'm all about customer choice in that respect. 00:10:25.440 --> 00:10:29.060 I'd like to make sure whatever customers are using that they get the best, you know, 00:10:29.060 --> 00:10:30.440 possible experience with our processors. 00:10:30.440 --> 00:10:39.160 And so it just feels like whether, you know, it's OpenStack work or some of the things that people are using Python for in, 00:10:39.240 --> 00:10:46.480 like high-performance computing or big data analytics, machine learning. 00:10:46.480 --> 00:10:49.300 There's just a lot of applications of Python. 00:10:49.300 --> 00:10:59.560 And it just seems like for Intel as a company to not be kind of investing that and making it great on our processors, 00:10:59.880 --> 00:11:01.700 it's just not a good idea. 00:11:01.700 --> 00:11:10.460 So one of the things I love about this is that we're being agile with respect to how we invest in the software that's running on the data center on our processors. 00:11:10.460 --> 00:11:16.080 And we're trying to, again, if our customers are using this stuff, we want to make sure that we have the best sort of, you know, 00:11:16.080 --> 00:11:17.360 the best sort of experience for them. 00:11:17.460 --> 00:11:22.300 So hopefully they'll come along and buy our processors on an ongoing basis. 00:11:22.300 --> 00:11:24.200 Yeah, I hope so. 00:11:24.200 --> 00:11:25.140 It sounds great. 00:11:25.140 --> 00:11:31.480 So can you give me some idea of what, like, what are you studying in Python? 00:11:31.480 --> 00:11:42.780 Are you looking at sort of the CPython implementation and how it's working and understanding how that's running on your chips and things like this? 00:11:42.780 --> 00:11:43.660 Or what's the story there? 00:11:43.660 --> 00:11:45.060 Yeah, yeah, yeah. 00:11:45.060 --> 00:11:52.100 But our philosophy is, you know, it's kind of what I call the core software strategy, which basically says this. 00:11:52.100 --> 00:12:00.700 It's like you can go off and, you know, work on speeding up some, you know, customer's application or some other, you know, kind of system. 00:12:00.700 --> 00:12:02.880 You know, that's great. 00:12:02.880 --> 00:12:07.920 But then, you know, at the end of the day, you know, you've optimized that one piece of software. 00:12:07.920 --> 00:12:11.720 You haven't really, you know, optimized anybody else's Python code, right? 00:12:11.720 --> 00:12:15.680 So, for example, let's take Swift. 00:12:15.680 --> 00:12:22.280 It's an open source object storage subsystem that's part of OpenStack, and it's implemented in Python. 00:12:22.280 --> 00:12:29.040 And if you look at that and you go, well, I could go work on the Swift source code, right, the Python source code to make it faster. 00:12:29.120 --> 00:12:34.560 But then it'd be better if I could actually speed up Python instead, that core software. 00:12:34.560 --> 00:12:41.640 And so, in theory, if I speed up Swift, I'll also speed up anything else that's using that accelerated Python, right? 00:12:42.020 --> 00:12:49.260 So, the way we achieve that core software strategy is we've developed some really amazing abilities. 00:12:49.260 --> 00:12:59.440 I mean, the engineers on the team have just done an amazing job to analyze exactly where the processor is spending its time as it's running, you know, major Python code, right? 00:12:59.440 --> 00:13:08.080 So, we're actually able to go at a very deep level, and we had some really very interesting discoveries that we've made at the microarchitectural level. 00:13:08.080 --> 00:13:10.880 And, you know, I can tell you more about that. 00:13:10.880 --> 00:13:12.820 It's kind of interesting. 00:13:13.100 --> 00:13:19.080 I mean, clearly, we see some things that are some challenges from a microarchitecture standpoint. 00:13:19.080 --> 00:13:34.160 So, we go, well, if we're going to improve the performance, we need to do what we can to address those things with the current products and then give feedback to our chip architects to say, hey, man, you need to change the way that the chip is designed to run these languages better. 00:13:34.160 --> 00:13:38.360 So, this is a – it's like a – not an either-or for us. 00:13:38.380 --> 00:13:39.300 It's like a both-and. 00:13:39.300 --> 00:13:49.520 Understanding the architecture at a deep level, we can go in and really speed up Python, and we can also go tell the chip architects to design better chips. 00:13:49.520 --> 00:13:56.900 So – and along those lines, yeah, our philosophy is totally upstreaming all of these things. 00:13:56.900 --> 00:14:04.700 And so, as much as possible, I sort of go – I mean, philosophically, I like to put the cookies on the low shelf so everyone can get at them. 00:14:04.780 --> 00:14:09.980 As much as possible, I want to make sure that we don't just do some amazing sort of things and that we upstream them. 00:14:09.980 --> 00:14:14.160 But then, you know, I want to make sure people know about them, right, so they can take advantage of them too. 00:14:14.160 --> 00:14:20.300 And I want to make it – I want to be a contributor to the community so that they can, you know, take advantage of the hard work that we're doing as well. 00:14:20.300 --> 00:14:22.280 Yeah, that's really excellent. 00:14:22.540 --> 00:14:24.780 So, it manifests sort of in two ways. 00:14:24.780 --> 00:14:31.140 You can improve CPython, and we're just getting started talking about the possibilities there, right? 00:14:31.140 --> 00:14:35.420 And then, how dramatic of a change can you make at the chip level? 00:14:35.420 --> 00:14:42.760 I mean, I think it's really cool that you're, like, thinking about, okay, well, here's our current architecture and the way things are working. 00:14:43.280 --> 00:14:46.300 But when you run Python code, this could be better. 00:14:46.300 --> 00:14:50.120 Would the chip actually detect what type of code it's running? 00:14:50.120 --> 00:14:52.200 Like, does it work at that level? 00:14:52.200 --> 00:14:56.060 Or are they just – you're like, these types of operations could be better? 00:14:56.440 --> 00:15:04.040 Well, actually, our chip architects have a long practice of looking at how current software runs on current chips. 00:15:04.040 --> 00:15:16.200 And we actually have some great tools to be able to analyze that at an instruction level that says, okay, you're running – oh, I don't know whether it's a database or whether it's, you know, Python or something else. 00:15:16.200 --> 00:15:19.840 And you're able to analyze, you know, okay, here's the series of instructions that get run. 00:15:19.840 --> 00:15:29.060 And then they can play a lot of interesting, you know, conceptual games with that and say, well, what if we added this instruction, a new instruction to optimize this thing? 00:15:29.060 --> 00:15:34.060 Or what if we change the way the cache works to, you know, run this stuff better? 00:15:34.060 --> 00:15:48.700 Or, you know, and, you know, we're actually – by having us involved with them, we're actually able to come up with some really interesting ideas and say, well, hey, maybe we could have this kind of, you know, acceleration 00:15:48.700 --> 00:15:53.280 or this kind of idea about how to run these languages better, right? 00:15:53.280 --> 00:16:05.100 So our hope is that, yeah, you know, and so the current chips are really designed from just years of running code, you know, of existing software against it. 00:16:05.100 --> 00:16:09.600 And so, you know, some of it's making up for bad software practices, bad programming practice. 00:16:09.600 --> 00:16:16.760 And it's like, oh, if people were to code this way, maybe if we organized the chip this way, it would run, you know, sort of poor code faster or something like that. 00:16:16.760 --> 00:16:18.260 Not to say that anyone's code is crap. 00:16:18.260 --> 00:16:18.980 I'm not saying that. 00:16:18.980 --> 00:16:20.960 But there's a lot of crappy code that's out there, right? 00:16:20.960 --> 00:16:21.260 Yeah. 00:16:21.260 --> 00:16:23.040 I'm sure none of your listeners have bad code. 00:16:23.040 --> 00:16:24.080 I'm sure they don't. 00:16:24.080 --> 00:16:26.820 But if they listen to other podcasts, they may. 00:16:26.820 --> 00:16:27.560 I'm just teasing. 00:16:28.080 --> 00:16:30.840 I'm sure they're fixing other people's bad code, not their own. 00:16:30.840 --> 00:16:35.620 But, I mean, you know, that's a key, you know, observation in terms of the chip. 00:16:35.620 --> 00:16:46.440 Now, I couldn't necessarily go into a lot of ideas on the podcast, but we're working on a bunch of really intriguing ideas, in my opinion, about ways that we can really run this stuff much, much better. 00:16:46.640 --> 00:16:59.660 I think it's cool to know that you guys out there are actually thinking about, you know, specifically studying these different runtimes like Python and Node.js and PHP and so on. 00:16:59.660 --> 00:17:06.380 And understanding sort of how they're working and not working on your hardware and then adjusting for that. 00:17:07.200 --> 00:17:11.240 When would the first sort of fruits of these labors show up? 00:17:11.240 --> 00:17:18.540 You said in 2010 you got started with the Linux embedded stuff and then later you got into Python. 00:17:18.540 --> 00:17:21.680 So how far down the road does this stuff land? 00:17:21.680 --> 00:17:22.860 Yeah. 00:17:22.860 --> 00:17:31.200 And by the way, the one thing I was going to add to your observation about the work that we do, we've actually been doing this work with Java now for, I think, 15 years. 00:17:31.200 --> 00:17:43.900 And we have this nice little chart that we compare from generation to generation how much the chip speeds it up, but how much our software, you know, JVM's, you know, changes have accelerated in each of those generations. 00:17:43.900 --> 00:17:48.780 And so we're just playing back the same playbook, basically, that we've been using for years with Java. 00:17:48.780 --> 00:18:00.560 What I would say in terms of the fruits of our labor, actually, when we began the group, we had our first performance patches available to the community, I think, within about like three, four months or so. 00:18:00.560 --> 00:18:03.720 And we're on the order of like a 10% performance boost. 00:18:03.720 --> 00:18:06.440 And, you know, you think, well, 10%, that can't be great. 00:18:06.440 --> 00:18:21.280 But, in fact, here's my experience with data center level code, whether it's, you know, massive database benchmarks like TPCC or this, you know, running Swift or some of these other big, big, big customer workloads on a data center. 00:18:21.640 --> 00:18:26.700 If you can boost something, you know, 5% to 10%, you are just like, you are like golden. 00:18:26.700 --> 00:18:27.840 That's like awesome. 00:18:27.840 --> 00:18:35.940 It's very unusual to find a real customer benchmark that gets, you know, some multiple of, you know, many X speed. 00:18:35.940 --> 00:18:37.220 It just doesn't happen, right? 00:18:37.440 --> 00:18:43.840 And so, realistically, if you can improve the throughput by a few percent, usually people are pretty happy about that. 00:18:43.920 --> 00:18:45.700 So, we came up with our first patches. 00:18:45.700 --> 00:19:03.300 Yeah, I'm pretty sure it was about three, four months after we started analyzing things and look at some low-hanging fruit of places where we could pull in some stuff pretty quickly and get the experience with the community, you know, talking with Guido, talking with other community developers to try and really, you know, hone our, you know, our efforts there. 00:19:04.080 --> 00:19:14.880 And the first time we kind of took a look at PyPy, that's when we said, well, shoot, maybe we could get a much, much higher, you know, return with PyPy. 00:19:14.880 --> 00:19:17.520 Maybe we could get a whole lot more than just a few percent here and there. 00:19:17.520 --> 00:19:20.140 So, that became extraordinarily interesting. 00:19:20.140 --> 00:19:20.140 Thank you. 00:19:20.140 --> 00:19:50.120 Thank you. 00:19:50.120 --> 00:20:20.100 Thank you. 00:20:20.100 --> 00:20:25.060 Yeah, that's awesome. 00:20:25.060 --> 00:20:27.440 And that's kind of what I was hinting at, like we're just getting started. 00:20:27.440 --> 00:20:32.580 So, the majority of people who run Python code, they run it on CPython. 00:20:32.580 --> 00:20:37.120 That's the default thing you get if you go to python.org and you download it. 00:20:37.120 --> 00:20:41.740 But there are many other runtimes or implementations. 00:20:41.740 --> 00:20:47.160 The most popular one around performance is probably one called PyPy, P-Y-P-Y. 00:20:47.160 --> 00:20:52.780 But we also have PyStin coming out of Dropbox and Guido's group there. 00:20:53.320 --> 00:20:56.840 We've got Psython, Jython, IronPython. 00:20:56.840 --> 00:20:58.840 There's a lot of options. 00:20:58.840 --> 00:21:07.440 And so, you turn to the probably one of the more established high-performance alternatives to CPython, PyPy, right? 00:21:08.060 --> 00:21:08.620 That's correct. 00:21:08.620 --> 00:21:08.620 That's correct. 00:21:08.620 --> 00:21:16.520 And, you know, for us, it's like I didn't want to be married to anything until we really got a chance to see, you know, kind of what was going on in the landscape. 00:21:16.980 --> 00:21:19.060 I mean, we looked at all of these, frankly. 00:21:19.060 --> 00:21:22.000 And they all have their pros and cons. 00:21:22.000 --> 00:21:26.500 I mean, things like Psython, et cetera, you know, basically creating C code. 00:21:26.500 --> 00:21:29.300 That's a nice model for performance. 00:21:29.300 --> 00:21:34.560 The challenge is that kind of takes away some of the, you know, development speed you get from an actual interpreted language. 00:21:35.320 --> 00:21:42.660 You know, and PyPy for it, and Pidgin and Piston have interesting qualities. 00:21:42.660 --> 00:21:46.920 I don't, you know, the thing that I like about PyPy is as follows. 00:21:46.920 --> 00:21:49.540 I mean, for one thing, it's been around 10 years, right? 00:21:49.540 --> 00:21:55.320 So, it's got, as far as I can tell, the broadest compatibility of any of these efforts. 00:21:56.020 --> 00:21:58.640 It's focused on both Python 2 and Python 3. 00:21:58.640 --> 00:22:11.480 I think that, you know, it really hasn't received a ton of, you know, broad sort of help from folks, you know, like these performance guys that I work with here at Intel. 00:22:11.480 --> 00:22:17.220 I mean, I think they've done an amazing job, absolutely amazing, stunning. 00:22:17.220 --> 00:22:19.160 I am incredibly impressed. 00:22:19.160 --> 00:22:33.840 We had a great sprint with them, you know, basically in March of this year, and face-to-face effort with their core developers, and it was just really very, very effective. 00:22:33.840 --> 00:22:38.980 And then there was another release of PyPy that came out as a result of the sprint. 00:22:38.980 --> 00:22:42.120 You know, so I think we're just getting started working with it. 00:22:42.120 --> 00:22:44.880 The results have been nothing short of stunning. 00:22:45.140 --> 00:22:51.620 I mean, this is what really impressed me because, like I said, we're working with Swift because it's a part of OpenStack. 00:22:51.620 --> 00:22:53.480 We're doing a lot of stuff with OpenStack at Intel. 00:22:53.480 --> 00:22:58.000 You know, Swift is the part of OpenStack that seems to spend a lot of its time in Python. 00:22:58.000 --> 00:23:06.960 As we analyzed it, 70% of the cycles, 70%, 80% of the cycles when you're running Swift is actually in the Python interpreter itself. 00:23:06.960 --> 00:23:10.760 So, it's like, oh, there's a lot of opportunity there. 00:23:11.220 --> 00:23:19.260 Then we look at, we split that down further, and it's like really being, you know, like 30% of those cycles are just in the main interpreter loop of CPython. 00:23:19.260 --> 00:23:24.160 So, we said, well, let's try PyPy, as you said, P-Y-P-Y, and it's incredible. 00:23:24.160 --> 00:23:26.820 We got 111% throughput improvement. 00:23:26.820 --> 00:23:36.120 Now, pause for a second and say, okay, I was looking at a few percent here and there, and here I've gotten more than double the throughput of using PyPy, right? 00:23:36.200 --> 00:23:38.740 It's like compared to CPython, amazing. 00:23:38.740 --> 00:23:42.440 And then it's like an 87% response time improvement. 00:23:42.440 --> 00:23:50.900 And it's like if something's like a software system, I mean, throughput's great because, right, you can scale up more and more users and you get, you know, great throughput. 00:23:50.900 --> 00:23:54.640 But the response time is what people really respond to, right? 00:23:54.660 --> 00:24:05.040 I mean, this is like it means they're accessing files, whether you're like, for example, Wikipedia, all of the, you know, sort of images that people look at in Wikipedia are all managed by Swift, right? 00:24:05.040 --> 00:24:09.160 So, you can look at something like that, and that just brings up your Wikipedia pages faster. 00:24:09.960 --> 00:24:14.640 And I've talked to various customers who are using Swift for their, you know, object storage system. 00:24:14.640 --> 00:24:18.540 And, yeah, this is a huge deal when you get that kind of improvement. 00:24:18.540 --> 00:24:33.620 So, it's like, well, with such amazing, you know, speedups, why wouldn't we try to, you know, see if we can, you know, provide a little, you know, love to the project to see if we can, you know, really do it, you know, continue to make this maybe even the default of how people use Python. 00:24:34.240 --> 00:24:36.320 Okay, yeah, that's really amazing. 00:24:36.320 --> 00:24:39.980 Do you know if you can run all of OpenStack on PyPy? 00:24:39.980 --> 00:24:44.000 Yeah, we're actually in the process of getting that together. 00:24:44.000 --> 00:24:51.000 We have, let's see, so far Swift, Keystone, Nova, and Neutron ported. 00:24:51.000 --> 00:24:59.300 And I would, you know, I would say trying to get like a proof of concept where we have all the core services running in PyPy. 00:25:00.000 --> 00:25:05.720 And I'd like us to really be able to do that maybe before the next OpenStack Summit in the fall in Barcelona. 00:25:05.720 --> 00:25:08.860 So, I'm hoping that we'll be able to show that off. 00:25:08.860 --> 00:25:12.720 Keystone, I said, is another interesting service that's written in Python. 00:25:12.720 --> 00:25:18.200 Keystone, if you're not familiar with it, is the user authentication part of OpenStack. 00:25:18.200 --> 00:25:26.660 So, essentially, every OpenStack service has to go through Keystone to see if you're authorized to do the things you're saying you want to do, right? 00:25:27.020 --> 00:25:30.160 And so, it's a really, you know, centralized part of the project. 00:25:30.160 --> 00:25:33.440 And we were able to speed it up by 37% using PyPy. 00:25:33.440 --> 00:25:36.660 So, that's, again, pretty darn amazing. 00:25:36.660 --> 00:25:42.120 So, I think I'd like to see the entire, yeah, I'd like to see PyPy as the default. 00:25:42.120 --> 00:25:48.380 In fact, years ago, the OpenStack gate had a requirement that everything had to work with PyPy. 00:25:48.500 --> 00:25:50.200 And a year or so ago, that got dropped. 00:25:50.200 --> 00:25:51.560 And I'd like to bring that back. 00:25:51.560 --> 00:25:58.940 And, if possible, try and drive, you know, something that would have the whole, you know, community basically making use of PyPy within OpenStack. 00:25:58.940 --> 00:25:59.200 Yeah. 00:25:59.200 --> 00:26:00.720 Yeah, that's really cool. 00:26:00.720 --> 00:26:04.740 If you're running something like OpenStack, you're running on a lot of machines. 00:26:04.740 --> 00:26:09.180 And efficiency is going to make a big difference in that type of system, right? 00:26:10.200 --> 00:26:10.600 Absolutely. 00:26:10.600 --> 00:26:14.980 And, you know, if you think about this, in some respect, it's not kind of radical. 00:26:14.980 --> 00:26:28.460 Every other interpreted language, whether it's Java, JavaScript, PHP, you name it, Lua, they have all gone the direction of a JIT in order to improve performance. 00:26:29.340 --> 00:26:33.340 And here, let me tell you how this works, right? 00:26:33.340 --> 00:26:36.620 If you've got two, let's say you're adding two integers together, right, in Python. 00:26:36.620 --> 00:26:37.600 Very simple operation. 00:26:37.600 --> 00:26:44.920 You know, if you're running native code on our processors, right, there's one instruction that adds two integers together, right? 00:26:44.920 --> 00:26:50.260 In some cases, it can be even less than one instruction with opcode fusing some features that we have. 00:26:50.620 --> 00:27:00.540 With Python, CPython in particular, we did a measurement and showed that on average, it takes 76 instructions to add two integers together, right? 00:27:00.540 --> 00:27:03.400 So automatically, you can see a stunning difference. 00:27:03.400 --> 00:27:05.200 And, you know, we've got great processors. 00:27:05.200 --> 00:27:07.320 They run instructions very fast. 00:27:07.320 --> 00:27:11.680 But you're running one instruction versus 76 is definitely going to be faster. 00:27:11.680 --> 00:27:14.280 That's a very hard thing to compensate for. 00:27:14.280 --> 00:27:15.520 Yeah. 00:27:15.740 --> 00:27:22.200 So as a result, you know, the processor spends a lot of it because the code footprint is just huge on something like Python. 00:27:22.200 --> 00:27:25.800 And so, you know, you spend a lot of the processor spends a lot of its time. 00:27:25.800 --> 00:27:29.860 We talk about it being cycles stalled in the front end. 00:27:29.860 --> 00:27:33.140 And what that basically means is the first two stages. 00:27:33.140 --> 00:27:37.620 If you think about it, I'm sorry to, let me just mention this briefly. 00:27:37.620 --> 00:27:38.960 I won't geek out on this too much. 00:27:39.040 --> 00:27:49.000 But it's like if you think about the five-stage pipeline for a processor, a modern processor, you get best performance if you can get all the stages of the pipeline running in parallel, right? 00:27:49.000 --> 00:27:55.660 The problem is our processors are spending like half of their time twiddling their thumbs waiting for new instructions to get fetched and decoded. 00:27:56.480 --> 00:28:09.060 So if we can, you know, make a huge difference to that, and a lot of it, frankly, comes down to simply the size of the instructions, the size of the code footprint that we're trying to process, right? 00:28:09.060 --> 00:28:20.120 And so JIT directly addresses that because what a JIT is going to do is it's going to look at the hot interpreted, you know, code and generate native code, right? 00:28:20.120 --> 00:28:28.180 So instead of that 76 instruction for an ad, if it's a hot ad, it will basically run in one instruction because it will be running native code instead. 00:28:28.180 --> 00:28:31.700 So this is not a huge secret. 00:28:31.700 --> 00:28:32.980 It's not magical. 00:28:32.980 --> 00:28:37.360 It's something that is a really, you know, well-worn technique to make this happen. 00:28:37.360 --> 00:28:41.540 And I'm telling you, every other language has gone this direction. 00:28:41.540 --> 00:28:46.940 And as I've, you know, as Guido and I have talked about this as well, he's like, you know, he's, you know, supportive. 00:28:46.940 --> 00:28:48.920 He's like, he says, I don't have a dog in the fight. 00:28:48.920 --> 00:28:54.960 You know, he's basically, he's more, I think, interested in the advances of the language as opposed to these performance things. 00:28:54.960 --> 00:28:57.240 Usually he's not thinking in terms of the performance side. 00:28:57.240 --> 00:28:59.200 But I'm, you know, I'm thinking about the performance side. 00:28:59.200 --> 00:29:11.100 I said, hey, could we really help the overall project by, you know, having a, you know, a great interpreter for Python that really runs really super fast? 00:29:11.100 --> 00:29:18.360 By the way, they're able to achieve greater performance by several other things in terms of the garbage collection that they're able to do. 00:29:18.360 --> 00:29:22.420 In various ways, they have much lower memory footprint as well. 00:29:22.420 --> 00:29:27.260 And so this is something that several of these techniques they're able to use to get really stunning performance. 00:29:27.840 --> 00:29:30.000 Yeah, the PyPy guys are doing a great job. 00:29:30.000 --> 00:29:32.520 And they definitely do some interesting stuff. 00:29:32.520 --> 00:29:34.540 I talked to them on episode 21. 00:29:34.540 --> 00:29:43.640 One of the things that I thought was interesting is the way that they don't immediately JIT everything, but they kind of wait until they find a hotspot. 00:29:43.640 --> 00:29:45.120 And then they go and JIT that. 00:29:45.380 --> 00:29:45.520 Right. 00:29:45.520 --> 00:29:46.020 Right. 00:29:46.020 --> 00:29:52.260 That's, again, a well-worn technique that has a ton of fruit in the Java space, as an example. 00:29:52.260 --> 00:29:52.540 Right. 00:29:52.540 --> 00:29:55.480 They even called their JIT a hotspot is what it's called. 00:29:55.480 --> 00:29:55.780 Right. 00:29:55.780 --> 00:29:59.620 So this is a common technique, and I think it works really well. 00:29:59.620 --> 00:30:00.600 Yeah. 00:30:00.600 --> 00:30:02.380 Yeah. 00:30:02.380 --> 00:30:13.700 I think the challenges that you run into is languages like Python, they're so flexible and you can change the types so much that it's not as easy as JIT in something like Java. 00:30:13.700 --> 00:30:14.020 Right. 00:30:15.020 --> 00:30:15.700 Oh, yeah. 00:30:15.700 --> 00:30:23.180 And particularly when, you know, because it's one of the really attractive things about the language is how, you know, flexible it can be in terms of typing. 00:30:23.180 --> 00:30:27.520 You don't have to spend a lot of time, you know, coming up with the types and documenting all those things. 00:30:27.520 --> 00:30:33.080 And so, you know, coding that is Pythonic, it really has that, you know, kind of power and flexibility. 00:30:33.080 --> 00:30:41.560 Unfortunately, that power and flexibility also comes with a cost and typically, you know, means that you have to spend a lot of code trying to figure out, now, what is the type of this thing? 00:30:41.560 --> 00:30:44.100 And did it change since the last time I looked at it? 00:30:44.100 --> 00:30:44.660 Right. 00:30:44.660 --> 00:30:48.480 So that's a strong consideration when trying to run this stuff fast. 00:30:48.480 --> 00:30:49.660 Yeah. 00:30:49.660 --> 00:30:54.540 I think there's a lot of attention being given to speed and Python lately. 00:30:54.660 --> 00:31:09.740 I've noticed a big uptick in projects and people focused on trying to make Python faster, be that through PyPy or the Microsoft Pigeon or I just spoke to the Euphoria guys about distributed, compiled Python. 00:31:10.020 --> 00:31:13.940 There's a whole bunch of really interesting things. 00:31:13.940 --> 00:31:16.240 And they're not all mutually exclusive. 00:31:16.240 --> 00:31:20.160 I think they're going to come together and some pretty amazing stuff is going to come out of it. 00:31:20.160 --> 00:31:21.500 Yeah, I agree with you. 00:31:21.580 --> 00:31:28.040 I think that one of the strengths of the Python community is the fact that you do have a lot of freedom for innovation. 00:31:28.040 --> 00:31:33.320 And, you know, the uptick in performance projects, I've noticed that as well. 00:31:33.320 --> 00:31:44.840 I think part of that's just because it has become so well adopted that I think if they don't solve the performance thing, you know, I think people will sort of get frustrated and go on to other things. 00:31:45.000 --> 00:31:55.920 You know, for example, in the high-performance computing space, you have people who are scientists and, you know, people studying data and things of that sort as opposed to being programmers, right? 00:31:55.920 --> 00:32:00.140 And they love Python because they can, you know, implement their code using Python. 00:32:00.140 --> 00:32:05.820 But then they go, well, this is performing slowly, so, you know, let me recode it in something else, right? 00:32:05.820 --> 00:32:13.340 And I think our vision would be that people don't have to, you know, change their language to get better performance. 00:32:13.480 --> 00:32:18.440 I'd love it to be a no-compromises, you know, experience with Python, right? 00:32:18.440 --> 00:32:26.480 And then we can bring to bear the best of our, you know, microarchitectural analysis tools, software optimization tools, and, frankly, analysis tools as well. 00:32:26.480 --> 00:32:38.740 We've got a terrific visual profiler called VTune that lets you, you know, pinpoint the exact area in your Python code that is, you know, causing your problems, performance problems. 00:32:38.840 --> 00:32:49.740 We also have, you know, a set of Python libraries that are, particularly if you're using NumPy, SciPy, Pandas, SciKit Learn, a variety of those packages. 00:32:50.300 --> 00:32:54.900 We have an accelerated version of those as well in a Python product. 00:32:54.900 --> 00:33:04.380 So there's a lot of, you know, but our common vision, you know, goes end-to-end in that sort of thing, whether it's the upstream stuff that we're doing or the Python product. 00:33:04.380 --> 00:33:10.600 We just want to make sure that the experience people have with Python is no compromises relative to performance. 00:33:11.180 --> 00:33:11.900 Yeah, that's great. 00:33:11.900 --> 00:33:15.800 And I want to talk about the NumPy, the data science stuff in a moment. 00:33:30.640 --> 00:33:37.840 This portion of Talk Python To Me is brought to you by Metis, offering data science training in New York City, Chicago, San Francisco, and online. 00:33:37.840 --> 00:33:51.240 Led by a deep team of senior data scientists, Metis delivers immersive boot camps, corporate workshops, online training, and part-time professional development courses in data visualization, machine learning, big data, and other data science skills. 00:33:51.240 --> 00:33:57.240 Their full-time boot camp is the only accredited data science boot camp available and includes extensive career support. 00:33:57.880 --> 00:34:08.240 Metis maintains a busy event schedule, so be sure to check them out on Meetup and keep in touch via at ThisIsMetis on Twitter and learn more about them on the web at ThisIsMetis.com. 00:34:08.240 --> 00:34:23.200 So you said that you had worked with the PyPy guys and you actually got them together to do a sprint with some people on your team. 00:34:23.200 --> 00:34:24.980 Is that what the story is? 00:34:24.980 --> 00:34:25.420 Yeah, that's right. 00:34:25.920 --> 00:34:35.280 Yeah, it turns out that all of the core PyPy developers are in Europe, except for Maciek, who's in South Africa. 00:34:35.280 --> 00:34:40.780 And I have a staff of developers working on Python that are in Europe. 00:34:40.780 --> 00:34:42.800 And so we brought everybody together. 00:34:42.800 --> 00:34:45.840 It turns out it's in Bucharest, Romania. 00:34:46.500 --> 00:34:51.140 And so we got everybody together physically in the same room. 00:34:51.140 --> 00:34:57.720 There were actually a couple of folks from the academic world, a couple of universities that wanted to come in and, you know, work. 00:34:57.720 --> 00:35:01.620 Once I got wind of this happening, they said, man, we want to get involved in this, too. 00:35:02.180 --> 00:35:08.940 So, you know, it was a very effective sprint and great to get that teamwork going. 00:35:08.940 --> 00:35:16.880 I feel like we now have a great understanding not only of, you know, PyPy, but also the microarchitectural tools that we have to analyze performance and things of that sort. 00:35:16.940 --> 00:35:23.260 So we can now turn this thing, I think, into a really powerful kind of collaboration with the PyPy project. 00:35:23.260 --> 00:35:24.280 Oh, that's great. 00:35:24.280 --> 00:35:37.920 And I think that's really great that you were able to help get everyone together because that's one of the real big challenges of these open source projects that are somewhat large is just physically getting the people together. 00:35:37.920 --> 00:35:42.040 So there might be people on projects that have never met before, right? 00:35:42.040 --> 00:35:43.500 Absolutely. 00:35:43.500 --> 00:35:49.820 And, you know, I tell you, I've been working in specifically open source projects since like the mid-2000s. 00:35:49.820 --> 00:35:55.340 And it's true that, you know, open source would never have been, you know, wildly successful without the Internet. 00:35:55.340 --> 00:36:09.400 But trust me, when you can get people physically face-to-face, you can get human-to-human contact in a physical, you know, same physical space, you eliminate a lot of inefficiencies and barriers and you get a lot of progress very quickly. 00:36:09.400 --> 00:36:18.880 And so as far as I'm concerned, that's a worthwhile sort of investment to get the smart people together in a room and work on real problems and produce code as a result. 00:36:18.880 --> 00:36:21.780 Yeah, even just getting to know people for a long weekend. 00:36:21.780 --> 00:36:24.500 That can last like a whole year's worth of goodwill from that, right? 00:36:24.500 --> 00:36:25.920 Absolutely. 00:36:25.920 --> 00:36:26.760 No question. 00:36:26.760 --> 00:36:27.740 Yeah, great. 00:36:27.740 --> 00:36:33.880 So that's a huge commitment back to the PyPy group and all of that is open source, of course. 00:36:33.880 --> 00:36:40.800 Have you done anything where you've taken some of this work and research and gotten it back into CPython? 00:36:40.800 --> 00:36:47.460 Yeah, we're trying to be not an either or for me, at least. 00:36:47.460 --> 00:36:52.620 It's like there's still, you know, adoption of PyPy is currently not very high. 00:36:53.500 --> 00:36:55.460 A lot more people are running code on CPython. 00:36:55.460 --> 00:36:57.920 So it's not like an either or for me. 00:36:57.920 --> 00:36:59.120 It's more like a both and. 00:36:59.120 --> 00:37:03.360 And the same thing goes for Python 2 versus Python 3. 00:37:03.360 --> 00:37:17.960 I mean, even though Python 3 is the place where Guido and the rest of the community, the development community, would like everybody just to be doing things on Python 3, reality is core services and OpenStack, a lot of other code is still in Python 2. 00:37:18.640 --> 00:37:24.780 So that's our commitment to the community is as we do performance work, we're doing it for both Python 2 and Python 3. 00:37:25.720 --> 00:37:30.000 And we're really trying to pull off a thing where we can do CPython and PyPy as well. 00:37:30.000 --> 00:37:41.140 So, you know, we have, I think there's some chemical issues that we can, like I said, give a little bit of love to in terms of making it much easier to use PyPy in many of these circumstances. 00:37:41.140 --> 00:37:52.600 So there's some engineering work along those lines as well that we're trying to invest in also to eliminate if there are any deployment issues or any issues with, you know, with the garbage collector or something like that. 00:37:52.640 --> 00:37:55.060 So we're really trying to, you know, really eliminate those. 00:37:55.060 --> 00:38:03.040 And there is a class of code that is challenged partly because PyPy is so fast. 00:38:03.040 --> 00:38:10.040 You know, this is the amazing thing to me is that we've encountered a couple of places where, you know, people have coded a timeout in Python. 00:38:10.460 --> 00:38:13.580 And, you know, the code wasn't exactly correct. 00:38:13.580 --> 00:38:23.080 And so when you switch to PyPy, which is multiple times faster, right, suddenly these timeouts fail because, you know, the code, you know, isn't exactly correct. 00:38:23.080 --> 00:38:27.240 Now, it's not necessarily something I can fix in PyPy. 00:38:27.240 --> 00:38:31.440 I mean, I'd love to figure that out somehow, but I'm not sure that's going to be possible. 00:38:31.440 --> 00:38:36.820 But, you know, we've seen this now in a couple of very interesting instances where people, you know, write their code. 00:38:36.820 --> 00:38:45.460 And, you know, it's, you know, suddenly when you speed it up a lot, it breaks because it made some assumptions about how slow it should be running. 00:38:45.460 --> 00:38:47.740 It got to the weight too quickly, huh? 00:38:47.740 --> 00:38:50.940 It's an odd problem to have. 00:38:50.940 --> 00:38:57.640 But, you know, that is, frankly, when you're going to do a drastic improvement in performance, you may find an issue like that that pops up. 00:38:57.640 --> 00:39:00.140 So in any case, that's something we've seen a little bit of. 00:39:00.140 --> 00:39:05.340 But hopefully people can fix their code because I think that's probably a good thing, right? 00:39:05.340 --> 00:39:06.180 Yeah, absolutely. 00:39:06.560 --> 00:39:07.820 That's obviously the fix. 00:39:07.820 --> 00:39:09.360 Like, hey, let's just not wait so long. 00:39:09.360 --> 00:39:13.920 Anyway, so it sounds to me like you have this multi-pronged approach. 00:39:13.920 --> 00:39:20.220 You're trying to make the processors faster by understanding how they run Python code, trying to make CPython faster, trying to make PyPy faster. 00:39:20.220 --> 00:39:31.660 Another one of the prongs on making things faster or addressing this Python performance story is around something you call the Intel distribution for Python. 00:39:31.660 --> 00:39:33.340 And you kind of hinted at that before, right? 00:39:33.340 --> 00:39:34.160 What's the story there? 00:39:34.820 --> 00:39:40.180 Yeah, and this is, again, we have sort of two-pronged strategy here. 00:39:40.180 --> 00:39:44.960 One strategy says let's do, you know, everything upstream in open source. 00:39:45.160 --> 00:39:51.600 And the other is, is there some way we can kind of pinpoint some of the key pinch points that people have? 00:39:52.420 --> 00:39:57.780 And then it may be that we have some code, you know, internally that will accelerate those things. 00:39:57.780 --> 00:40:05.900 So, you know, in particular when you're running, you know, math functions, this is something that in the evolution of people's Python code, right? 00:40:05.900 --> 00:40:12.640 They write these things initially in Python, and then it's like, wow, this is kind of a performance pinch point. 00:40:12.640 --> 00:40:14.820 Is there something we can kind of do to improve it? 00:40:14.820 --> 00:40:20.920 So a lot of times people recode these things in C, and it's like, okay, and NumPy is a great example of that, right? 00:40:20.920 --> 00:40:25.360 So what we've actually been able to do is say, gee, we have this library here. 00:40:25.440 --> 00:40:26.380 It's called MKL. 00:40:26.380 --> 00:40:28.560 It's called our math kernel libraries, right? 00:40:28.560 --> 00:40:38.840 Where some incredibly smart people have gone at the deepest possible level to have math functions and execute them just as fast as possible, right? 00:40:38.840 --> 00:40:45.940 So they're highly tuned, very, you know, very, very bespoke sort of code, right? 00:40:45.940 --> 00:40:55.140 So, you know, what they've done is done an amazing job hooking up MKL with NumPy, SciPy, and these other popular libraries. 00:40:55.720 --> 00:41:00.480 And so that's code that they actually call the Intel Python distribution or Intel distribution. 00:41:00.480 --> 00:41:07.260 I might have the name wrong, but it's actually in beta now as we're talking, and so that should be something, 00:41:07.260 --> 00:41:12.920 some get in on the beta program, or if they're listening to this at a later time, they may, in fact, be already released. 00:41:12.920 --> 00:41:18.920 So that's another, I think, great alternative, particularly if you're using those scientific libraries to basically make use of that 00:41:18.920 --> 00:41:21.260 and see, you know, if that's going to improve things. 00:41:21.260 --> 00:41:27.540 In some cases, I think they found some code that's, you know, again, it can multithread some of these operations, 00:41:27.540 --> 00:41:31.960 whereas with, you know, regular Python, it's more or less single-threaded. 00:41:31.960 --> 00:41:35.180 And so with multithreading, depending on how many cores you can throw at it, 00:41:35.180 --> 00:41:39.080 they've seen up to 100x on some math functions that they've been able to speed up. 00:41:39.080 --> 00:41:40.740 So that's pretty stunning. 00:41:40.940 --> 00:41:47.840 There's also some data analytics functions that are part of this, and so I think that's a really intriguing option as well. 00:41:47.840 --> 00:41:49.980 Yeah, that does sound intriguing. 00:41:49.980 --> 00:41:56.620 And it sounds to me like what you're actually targeting there is you're targeting the C extensions 00:41:56.620 --> 00:42:02.280 or the C foundations of some of these heavily C-sped-up libraries. 00:42:03.320 --> 00:42:04.800 Yeah, that's exactly right. 00:42:04.800 --> 00:42:12.500 It's a strategy that, again, is like we don't want to have people make a choice between, well, do I want to stick with Python 00:42:12.500 --> 00:42:14.140 or do I want good performance, right? 00:42:14.140 --> 00:42:17.400 We want to be able to make it so that it's a no-compromise situation. 00:42:17.400 --> 00:42:24.140 So, frankly, this is another alternative that lets us, you know, target, you know, some users that really want super high performance 00:42:24.140 --> 00:42:25.620 in this mathematical area. 00:42:25.800 --> 00:42:30.200 And high-performance computing is another area that we've invested in pretty heavily here. 00:42:30.200 --> 00:42:33.340 So this is a way to, you know, get directly into that space and help them out. 00:42:33.340 --> 00:42:40.680 Yeah, and I guess you guys probably understand better than anybody how to exactly line up the math to work on your processors, right? 00:42:40.680 --> 00:42:46.920 Yeah, we really have some amazing tools and analytics to do. 00:42:46.920 --> 00:42:56.720 Yeah, when you sit next to the guys designing the chips, it's usually, you know, much easier to be able to squeeze out that last little bit of performance in any possible case. 00:42:56.720 --> 00:42:57.560 Yeah, yeah. 00:42:57.560 --> 00:42:58.080 Sounds great. 00:42:58.080 --> 00:43:00.000 I'll be sure to link to that in the show notes. 00:43:00.000 --> 00:43:06.440 So one of the things that we talked about last week is you were actually at the OpenStack conference, right? 00:43:06.440 --> 00:43:07.920 Maybe you can give us a... 00:43:07.920 --> 00:43:07.920 That's correct. 00:43:07.920 --> 00:43:08.660 OpenStack Summit. 00:43:08.660 --> 00:43:09.000 Yeah. 00:43:09.000 --> 00:43:09.360 Awesome. 00:43:09.360 --> 00:43:10.740 Give us a rundown on what happened there. 00:43:10.740 --> 00:43:11.620 Yeah. 00:43:11.880 --> 00:43:15.860 This is really exciting because the OpenStack Summit is, you know, every six months. 00:43:15.860 --> 00:43:18.800 This one was in Austin, Texas. 00:43:18.800 --> 00:43:32.120 And what was interesting about this is that OpenStack as a project is going through a lot of really interesting evolution as cloud computing obviously takes over the kind of the fastest growing segment of the data center kind of area, right? 00:43:32.120 --> 00:43:35.780 So software-defined infrastructure is incredibly important. 00:43:35.780 --> 00:43:39.960 And something like 70% of OpenStack is actually written in Python. 00:43:40.160 --> 00:43:51.340 So one of the things that we did was jointly with the project technical leader, or PTL, of Swift, he and I have been collaborating together on PyPy, 00:43:51.340 --> 00:44:00.860 and we were able to jointly present our performance results and showing, hey, here's the effect performance-wise of using PyPy with Swift. 00:44:01.400 --> 00:44:03.820 And again, I think I mentioned the statistics earlier. 00:44:03.820 --> 00:44:05.840 It's like more than double the throughput, right? 00:44:05.840 --> 00:44:14.100 We have data which shows that we can, you know, get 111% throughput improvement and 87% response time improvement. 00:44:14.100 --> 00:44:16.420 You know, that's stuff that we've seen. 00:44:16.420 --> 00:44:20.260 The PTL, John Dickinson, works for a company called SwiftStack. 00:44:20.720 --> 00:44:25.480 And in their product sort of environment, they have a Swift-based product family. 00:44:25.480 --> 00:44:28.340 And so, yeah, they were able to see, you know, similar results. 00:44:28.340 --> 00:44:31.260 And so jointly we were able to present this. 00:44:31.260 --> 00:44:33.720 And people were very excited about this. 00:44:33.720 --> 00:44:39.760 The response was extremely positive, partly because of the opportunity to see the collaboration go on. 00:44:39.760 --> 00:44:47.000 You know, anytime you go in with open source and you can collaborate and talk about how we're working together to make people's lives better, 00:44:47.100 --> 00:44:48.820 I think it's all good, right? 00:44:48.820 --> 00:44:54.600 So that was – but what was funny is I've got to tell you a story. 00:44:54.600 --> 00:44:59.120 When I landed in Austin, so I get to my hotel that I'm staying at for this conference, 00:44:59.120 --> 00:45:02.420 and I'm walking over to the conference venue to get my conference badge. 00:45:02.420 --> 00:45:08.080 The first person I see on the street, I'm sad to say I didn't recognize him, 00:45:08.080 --> 00:45:11.040 but he recognized me and he said, oh, it's Mr. PyPy. 00:45:13.040 --> 00:45:18.780 Somehow, then I found out, okay, why have you recognized me? 00:45:18.780 --> 00:45:24.000 Anyway, so it's – I think the word's kind of getting out that this is something that, you know, 00:45:24.000 --> 00:45:25.080 we see some real value in. 00:45:25.080 --> 00:45:28.500 We'd like to try and help people out through advancing it a little bit. 00:45:28.500 --> 00:45:28.820 Yeah. 00:45:28.820 --> 00:45:34.220 Yeah, you must be being effective if you're getting recognized on the street as Mr. PyPy. 00:45:34.220 --> 00:45:34.680 That's awesome. 00:45:35.720 --> 00:45:41.480 My usual comment is a good reputation takes a long time to build, but a bad reputation is instantaneous. 00:45:41.480 --> 00:45:46.780 So, you know, I think we'll – you know, hopefully we can keep it to be a good reputation as opposed to a bad one. 00:45:46.780 --> 00:45:47.940 Yeah, absolutely. 00:45:47.940 --> 00:45:50.740 Okay, well, that sounds really interesting. 00:45:50.740 --> 00:45:53.400 That's great you were able to share your results with everyone. 00:45:54.520 --> 00:45:58.960 Another thing that you said that was pretty interesting, sort of completely unrelated to this, 00:45:58.960 --> 00:46:05.740 but you've been talking about how the Python community is a really welcoming community. 00:46:05.740 --> 00:46:12.260 And I can tell you from doing training and speaking at conferences that are not just Python conferences 00:46:12.260 --> 00:46:16.880 but have many languages and technologies that when I do Python talks, 00:46:16.880 --> 00:46:22.560 the group of people in the room are more representative of the group of people out in society, 00:46:22.560 --> 00:46:24.020 which is really great. 00:46:24.020 --> 00:46:25.920 Well, I totally agree with you. 00:46:25.920 --> 00:46:31.600 I think it's amazing to see, particularly because I've spent a number of years in other open source projects. 00:46:31.600 --> 00:46:42.420 And so I know at sometimes a painful level how challenging it can be for some people who are not males in particular. 00:46:42.420 --> 00:46:46.180 And this is one of the cases where I don't want to point to particular projects or something like that, 00:46:46.180 --> 00:46:48.000 but I've been involved in a number of projects. 00:46:48.000 --> 00:46:55.880 And it's hard sometimes to find projects that not only have great diversity but inclusion as well. 00:46:55.880 --> 00:46:59.160 And so one of the things that's been very impressive to me, 00:46:59.160 --> 00:47:02.800 the first Python conference I went to actually was EuroPython. 00:47:02.800 --> 00:47:08.200 And Guido made a – actually it was before Guido came on as a keynote speaker. 00:47:08.200 --> 00:47:11.660 The very first keynote of the project, of the conference, 00:47:11.840 --> 00:47:15.920 was a conference given by the founders of Django Girls. 00:47:15.920 --> 00:47:20.540 And it's two female engineers. 00:47:20.540 --> 00:47:28.980 And Django Girls was set up by then as a nonprofit to basically have women teach other women how to code in Python, right? 00:47:28.980 --> 00:47:30.600 And it was a great keynote. 00:47:30.800 --> 00:47:36.020 I mean they actually did a bunch of it as like a fable, you know, kind of talking about – 00:47:36.020 --> 00:47:39.700 essentially talking about how challenging it is for, you know, 00:47:39.700 --> 00:47:44.340 women to really be considered part of the community, right, of any sort of community. 00:47:44.340 --> 00:47:48.660 And so Django Girls, what's amazing is they started up with pretty much nothing. 00:47:48.660 --> 00:47:54.280 And after the first year, they had had 100 workshops all over the world. 00:47:54.480 --> 00:47:57.280 Again, women teaching women how to program in Python, right? 00:47:57.280 --> 00:47:59.780 And this was incredibly impressive. 00:47:59.780 --> 00:48:06.060 I mean Guido actually in his keynote, he – you know, I think he's part of the reason. 00:48:06.060 --> 00:48:07.240 It's like this. 00:48:07.240 --> 00:48:09.580 I think he's very committed. 00:48:09.580 --> 00:48:13.000 He was wearing a PyLadies T-shirt in the conference. 00:48:13.000 --> 00:48:19.520 PyLadies is another, you know, part of the Python community that works really well to help encourage women be involved in Python. 00:48:19.980 --> 00:48:23.320 He – in his keynote, he said – most of it was Q&A from the audience. 00:48:23.320 --> 00:48:26.840 And he said, let's alternate a man and a woman asking questions, right? 00:48:26.840 --> 00:48:30.520 So he made it super clear that that's a priority for him. 00:48:30.520 --> 00:48:31.460 And I asked him about it. 00:48:31.460 --> 00:48:34.220 Is that, you know, something that he – it's important to him? 00:48:34.220 --> 00:48:35.080 He said, yeah, absolutely. 00:48:35.080 --> 00:48:36.680 And so he's making a difference. 00:48:36.680 --> 00:48:39.440 He's putting his own, you know, kind of stamp behind that. 00:48:39.440 --> 00:48:50.640 And I might – on a personal basis, I had – you know, my daughter is – one of my daughters is 24 and she's still trying to find kind of her career direction. 00:48:50.640 --> 00:48:54.840 And she said, Dad, you know, if I wanted to learn programming, you know, what would you suggest? 00:48:54.840 --> 00:48:58.280 And I said, well, I think you should try Python. 00:48:58.280 --> 00:49:02.880 And I said, in fact, you know, you ought to check out this Django Girls thing to see if there's a workshop. 00:49:03.000 --> 00:49:05.900 Well, it turned out, you know, we live in the Portland area. 00:49:05.900 --> 00:49:09.200 There was a Django Girls workshop in another, like, six weeks. 00:49:09.200 --> 00:49:11.640 And I said, oh, look, you could get involved in that. 00:49:11.640 --> 00:49:20.160 And she submitted an application, explained, you know, what she'd be using her knowledge for and how – you know, why she wanted to learn it. 00:49:20.160 --> 00:49:21.760 They accepted her in the workshop. 00:49:21.760 --> 00:49:24.040 And, you know, she was a little nervous. 00:49:24.040 --> 00:49:28.300 She said, gee, I'm not sure I'm smart enough to actually, you know, do well with this, right? 00:49:28.300 --> 00:49:29.080 And I said, no, no, no. 00:49:29.080 --> 00:49:32.460 You know, and I understand sometimes people can feel that way. 00:49:32.500 --> 00:49:42.920 Typically, they've been made to feel not smart, which, you know, I try and do everything I can with my daughters to make them feel like they're – you know, they're both brilliant. 00:49:42.920 --> 00:49:45.000 They have no worry about that. 00:49:45.000 --> 00:49:51.320 But sometimes the environment is one that wants to, you know, make people who are, you know, women not necessarily feel very smart. 00:49:51.320 --> 00:49:53.940 And so I said, you know, I encouraged her. 00:49:53.940 --> 00:49:55.220 She went through the workshop. 00:49:55.220 --> 00:49:56.700 She was very successful with it. 00:49:56.940 --> 00:50:03.300 And I'm hoping, you know, fingers crossed that it would be something she'd be able to, you know, make a living doing if she digs it. 00:50:03.300 --> 00:50:08.220 So, yeah, I love the fact that Django Girls is – you know, has this available. 00:50:08.220 --> 00:50:12.560 And it's, again, part of – I think a core part of the Python community. 00:50:12.560 --> 00:50:20.600 I think it's one of the things that makes it a very attractive community is how important it is for the community to – for women to help, you know, other women. 00:50:20.600 --> 00:50:23.380 And for the entire community to have better inclusion. 00:50:23.380 --> 00:50:25.040 Now, is it perfect? 00:50:25.040 --> 00:50:25.720 No. 00:50:25.720 --> 00:50:28.460 I think there's some challenges still. 00:50:28.460 --> 00:50:33.780 I interviewed one of the – a fellow of the Python Foundation. 00:50:33.780 --> 00:50:36.460 Her name is Terry Oda. 00:50:36.460 --> 00:50:42.660 And I talked to Terry and I said – I asked her how it is from her perspective as a woman. 00:50:42.660 --> 00:50:46.120 I mean, it's one thing for me as a male to say, gee, this seems really good. 00:50:46.180 --> 00:50:47.800 Well, how about a perspective of a woman? 00:50:47.800 --> 00:50:52.040 And she's part of the, you know, the Intel – or, sorry, the Python Foundation. 00:50:52.040 --> 00:50:53.400 She's also an Intel employee. 00:50:53.400 --> 00:50:57.600 And she said, yeah, you know, she said that it's a good environment. 00:50:57.600 --> 00:51:03.420 It has the – you know, it could do better in terms of having more core developers who are women. 00:51:03.420 --> 00:51:11.300 And so there's more, you know, work that can be done to, you know, get more of that inclusion and diversity within, you know, sort of the core development community. 00:51:11.300 --> 00:51:15.860 But it is a great project for its diversity and inclusion. 00:51:15.860 --> 00:51:16.920 Yeah, that's cool. 00:51:16.920 --> 00:51:21.140 I think the Python community is more welcoming than most. 00:51:21.140 --> 00:51:23.920 And I think this is just, you know, more evidence of it. 00:51:23.920 --> 00:51:24.660 So that's cool. 00:51:24.660 --> 00:51:30.680 Hopefully we can inspire some more people to go out and check out some of these projects that are out there and freely available. 00:51:30.680 --> 00:51:31.440 Very cool. 00:51:31.440 --> 00:51:32.900 I agree, yeah. 00:51:32.900 --> 00:51:40.980 So getting near the end of the show, let me ask you a couple of questions that I always ask my guests while I still have you. 00:51:40.980 --> 00:51:44.280 If you're going to write some Python code, what editor do you open up? 00:51:44.720 --> 00:51:45.640 Oh, yeah. 00:51:45.640 --> 00:51:47.320 I'm an old VI guy. 00:51:47.320 --> 00:51:53.000 So, you know, with my roots back in Linux and back in the old days, or Unix, you know, before there was a Linux. 00:51:53.000 --> 00:51:55.500 So, yes, when I edit, I use VI. 00:51:55.500 --> 00:51:56.680 Okay, excellent. 00:51:56.680 --> 00:51:58.000 Yeah, that's definitely a popular one. 00:51:58.000 --> 00:52:04.960 And of all the PyPI packages, you can include Intel ones if you want, there's a ton of them. 00:52:04.960 --> 00:52:07.600 And nobody could possibly know about them all. 00:52:07.600 --> 00:52:10.040 There's, you know, close to 80,000, maybe over. 00:52:10.120 --> 00:52:10.900 I haven't looked in a while. 00:52:10.900 --> 00:52:17.420 What ones do you like that maybe you'd recommend people check out that are not necessarily the most popular? 00:52:17.420 --> 00:52:20.940 I'd ask the question, though, what's my favorite interpreter? 00:52:20.940 --> 00:52:24.840 My favorite interpreter for Python these days is definitely PyPy. 00:52:24.840 --> 00:52:25.980 Yeah, okay. 00:52:25.980 --> 00:52:26.820 That's great. 00:52:26.820 --> 00:52:28.640 It's a very meta answer, right? 00:52:28.720 --> 00:52:31.520 Like, it's PyPy that runs all the packages. 00:52:31.520 --> 00:52:32.600 I'm not a politician. 00:52:32.600 --> 00:52:34.640 I don't even get paid to be a politician. 00:52:34.640 --> 00:52:37.800 But sometimes I know how to, you know, reframe the question. 00:52:37.800 --> 00:52:38.700 Yeah, absolutely. 00:52:38.700 --> 00:52:44.980 Okay, so do you have a final call to action for listeners, things they should do or check out? 00:52:46.080 --> 00:52:51.800 Yeah, I would absolutely have people take another look at PyPy. 00:52:51.800 --> 00:52:53.320 You probably have in the past. 00:52:53.320 --> 00:52:59.040 In particular, we would love to have people check out, if you're using, you know, OpenStack or Swift, 00:52:59.040 --> 00:53:03.880 we would love to get people to, you know, check it out with the latest version of PyPy 00:53:03.880 --> 00:53:10.480 and really try to give us some feedback in the Swift community or in the Python dev community. 00:53:10.480 --> 00:53:12.080 We monitor Python dev. 00:53:12.080 --> 00:53:16.720 And if you have some observations about PyPy that would be helpful, saying, you know, for example, 00:53:16.720 --> 00:53:22.120 if you have some Python code and if you either find you have some, you know, challenges with it, 00:53:22.120 --> 00:53:26.440 we would love to hear that because one of the things we'd love to do is I've got engineers I can, you know, 00:53:26.440 --> 00:53:29.240 try and ask to, you know, basically say, hey, let's go solve these. 00:53:29.240 --> 00:53:34.100 I've given the call to action to our engineers to say, try and, you know, resolve these, 00:53:34.100 --> 00:53:39.240 any sort of compatibility issues or other deployment issues and let's make, you know, 00:53:39.240 --> 00:53:40.780 try and make this a great experience for people. 00:53:40.780 --> 00:53:45.960 So if you can help us out in terms of, you know, get your hands on PyPy, try it against your Python code 00:53:45.960 --> 00:53:50.740 and give us feedback in terms of what might be missing or could work better, we would love to hear that 00:53:50.740 --> 00:53:52.440 and we'd love to dialogue with you on it. 00:53:52.440 --> 00:53:53.100 Okay. 00:53:53.100 --> 00:53:57.220 So absolutely try your code on PyPy and send these guys some feedback. 00:53:57.220 --> 00:53:57.920 That's great. 00:53:58.880 --> 00:54:02.180 All right, David, it's been a super interesting conversation. 00:54:02.180 --> 00:54:08.440 I'm really excited to see the work that you're doing appear in the Intel chips going forward 00:54:08.440 --> 00:54:12.780 and in both the runtimes in CPython and PyPy. 00:54:12.780 --> 00:54:14.060 Thank you, Michael. 00:54:14.060 --> 00:54:16.100 I love talking about this stuff. 00:54:16.100 --> 00:54:20.800 I'm incredibly passionate about it and I'd love to, you know, 00:54:20.800 --> 00:54:25.140 I'd see our love for technology be able to be made use for a lot of people. 00:54:25.320 --> 00:54:29.060 So I'm hopeful that people will get excited about this stuff as a result too. 00:54:29.060 --> 00:54:30.020 Yeah, absolutely. 00:54:30.020 --> 00:54:31.240 Thanks for being on the show. 00:54:31.240 --> 00:54:31.700 Talk to you later. 00:54:31.700 --> 00:54:32.500 Thank you. 00:54:34.880 --> 00:54:48.860 Snap.CI is effective in deep training for data scientists. 00:54:48.860 --> 00:55:00.960 Metis is effective in deep training for data scientists. 00:55:00.960 --> 00:55:04.180 With immersive boot camps, in-person and online courses, 00:55:04.180 --> 00:55:08.420 you can learn the core data science skills you need to take your career to the next level. 00:55:08.420 --> 00:55:11.400 Are you or a colleague trying to learn Python? 00:55:11.400 --> 00:55:16.000 Have you tried books and videos that just left you bored by covering topics point by point? 00:55:16.000 --> 00:55:21.080 Well, check out my online Python course called Python Jumpstart by Building 10 Apps at 00:55:21.080 --> 00:55:25.740 talkpython.fm/course to experience a more engaging way to learn Python. 00:55:26.400 --> 00:55:31.640 You can find the links from this episode at talkpython.fm/episodes slash show slash 57. 00:55:31.640 --> 00:55:33.860 And be sure to subscribe to the show. 00:55:33.860 --> 00:55:36.180 Open your favorite podcatcher and search for Python. 00:55:36.180 --> 00:55:37.300 We should be right at the top. 00:55:37.300 --> 00:55:40.500 You can also find the iTunes feed at /itunes, 00:55:40.500 --> 00:55:43.260 Google Play feed at /play, 00:55:43.260 --> 00:55:47.720 and direct RSS feed at /rss on talkpython.fm. 00:55:48.220 --> 00:55:51.320 And there's been a few requests for the Ogthera format, 00:55:51.320 --> 00:55:53.480 a more open format than MP3. 00:55:53.480 --> 00:55:56.460 And I added that link as well in the footer of the website. 00:55:56.460 --> 00:56:00.060 Our theme music is Developers, Developers, Developers by Corey Smith, 00:56:00.060 --> 00:56:01.020 who goes by Smix. 00:56:01.020 --> 00:56:04.820 You can hear the entire song on talkpython.fm/music. 00:56:04.820 --> 00:56:07.080 This is your host, Michael Kennedy. 00:56:07.080 --> 00:56:08.700 Thank you so much for listening. 00:56:08.700 --> 00:56:11.000 Smix takes us out of here. 00:56:11.000 --> 00:56:15.100 Stating with my voice, there's no norm that I can feel within. 00:56:15.100 --> 00:56:17.940 Haven't been sleeping, I've been using lots of rest. 00:56:17.940 --> 00:56:20.800 I'll pass the mic back to who rocked his best. 00:56:20.800 --> 00:56:22.560 First developers. 00:56:22.560 --> 00:56:22.700 First developers. 00:56:22.700 --> 00:56:25.460 First developers. 00:56:25.460 --> 00:56:26.280 First developers. 00:56:26.280 --> 00:56:27.460 First developers. 00:56:27.460 --> 00:56:28.520 First developers. 00:56:28.520 --> 00:56:29.520 First developers. 00:56:29.520 --> 00:56:31.520 First developers. 00:56:31.520 --> 00:56:33.120 First of all, first of all, first of all, first of all. 00:56:33.120 --> 00:57:03.100 Thank you.