text stringlengths 0 2.67k |
|---|
**Jon Johnson:** Exactly. And the crucial bit for me is that I can use this technique on all existing container images that have ever been built, and it works. And so that's great. Because for my purposes, I want you to be able to look at any image quickly. And you can. |
\[01:05:52.11\]"The size of the index ends up being -- it depends, but heuristically, if you decide "Every four megabytes I'm going to take a checkpoint", then it works out to be about 1% the size of the original archive, because you're paying for 32 kilobytes for every four megs. |
For really big images, that 1% actually ends up being big still... And so the clever thing I did, I think, as a contribution to this whole idea, is that my index format, I made also a tar-gz file. And so if I decided that that index is too big for me to hold in memory, I reindex it. And so that thing itself is indexed.... |
**Justin Garrison:** Right. But even like a five-terabyte, you shrink that to 1% for your index, and then shrink that to 1% again, for your index of index... And you're not downloading a five-terabyte image to get the files you need. |
**Jon Johnson:** Yeah. This was my eureka moment during the symphony. I was like "Oh, I know what to do...!" |
**Justin Garrison:** \[laughs\] And that applies beyond just containers. I mean, you're obviously doing it with containers now, but even for things -- when I think of large language models, if we are compressing those, and we want to like decompress pieces of them, or if we only want to get like some portion of any lar... |
**Jon Johnson:** Yeah. And in fact, something that's interesting about that is that certain projects store their large language models as OCI things. So the Ollama stuff, you can look at it using my tools, because they're just in a registry somewhere. The other application of this, as we talked about at the beginning, ... |
**Autumn Nash:** You should be. |
**Justin Garrison:** Yeah, that's awesome. Anything that makes debugging easier - all for it. |
**Justin Garrison:** And faster, right? Same thing with like when you're marshalling and unmarshalling JSON. Like hey, what if we could speed up how fast this disk is, and just get past the problem by making that as extremely fast as possible? And in this case, what if we can get beyond the "I have to download this big... |
**Autumn Nash:** I feel like those are the hardest parts, too. Those are the things that you spend the most time on. Debugging, and trying to find things. If you can make that easier, you've already cut a lot of your engineering hours. |
**Jon Johnson:** Yeah. But what I do now is if I'm ever doing something where I'm just twiddling my thumbs, waiting on something really slow, it's like "I should spend this time making it not slow." And so anytime I'm waiting on CI, or whatever, for something, I'm like "Let's whip the profiler out and actually see what... |
**Justin Garrison:** Well, Jon, thank you so much for coming on and telling us all about RFC 1951, and how Deflate and Gzips work. That's just awesome. I love all the nitty-gritty in that, of how broadly that applies, and how specifically it allows things like your DAG to work, and to be performant, and to let people h... |
**Autumn Nash:** Can we get the poem, though? We need the poem. |
**Justin Garrison:** Oh, that's a Plus Plus thing, right? |
**Jon Johnson:** It could be. We could put a teaser in... So no joke, it's like a 10-minute read. |
**Autumn Nash:** Dang, you wrote a poem-poem. |
**Jon Johnson:** Yeah, I'm kind of embarrassed. I think maybe it should be Plus Plus subscriber content, but -- |
**Justin Garrison:** No, that's fine. If you are a Plus Plus subscriber and you want to hear the poem, Jon can read it out for us... If you're feeling comfortable. |
**Jon Johnson:** Well, I feel like I kind of owe it to you, because I spoiled -- |
**Justin Garrison:** And you did tease it, so... |
**Jon Johnson:** ...Jason's episode, you know... Let me take a sip of water, because this is -- it's gonna go \[unintelligible 01:10:07.15\] |
**Autumn Nash:** I'm really excited about this. |
**Break**: \[01:10:14.02\] |
**Justin Garrison:** Thank you so much, Jon, for coming on and sharing everything about tars, and Deflate, and all the things that I learned through this. It was fantastic. I knew about Jon's container websites before, to explore layers and containers, and I -- |
**Autumn Nash:** I wanna go back and listen to it, so I can like -- you know when you're just like "This is so cool, but I can't process all of it." I need to listen to it like three more times, and then draw some pictures and stuff, and then go check his website, and I'm gonna know all the things about tars. |
**Justin Garrison:** Yeah, getting hands-on with some of the tools would definitely help. |
**Autumn Nash:** Yeah. |
**Justin Garrison:** But for the outro today, I have like a weird -- I don't wanna say a story, but a history lesson maybe. Because something that we were talking about in the interview reminded me of something that I read in a book. And this book was called "A mind at play", and it was about Claude Shannon. And Claude... |
**Autumn Nash:** That is super-interesting. |
**Justin Garrison:** It was really -- like, when I read that, I'm like "Oh, my gosh." That is the simplest way that I can think about compressing something and encrypting something, where encrypting - you're making noise out of structure, and compression, you're trying to like take out all of that structure and make su... |
**Autumn Nash:** I just had like a whole epiphany about encryption... To give context, okay? ...this is how me and Justin's relationship grows. I get really excited about something, and then I text it to him, and then he's like "I read this book", and then he tells me about it... And at some point, both of our eyes lig... |
**Justin Garrison:** Right. Compression and encryption are just opposites of each other. And so let me walk through just a couple of his major things that he did in his life, because he was such a fascinating person. And the book, "A mind at play", it was really good, because it starts out as basically like -- |
**Autumn Nash:** Do you read audiobooks, or do you actually read these? |
**Justin Garrison:** I do audiobooks, yes. I can't sit down and look at words to consume the words. I consume them with my ears, not my eyes. |
**Autumn Nash:** It's good that you \[unintelligible 01:14:09.10\] That's really important in learning. |
**Justin Garrison:** He was a very eccentric person in a lot of ways, where he was a master unicyclist, he was a great juggler, and he also would have trumpets that would blow fire. He had all these toys and things that he would like to play with and explore and do... And he worked at Bell Labs, and he was kind of know... |
**Autumn Nash:** Okay, low-key though, is this not the best buy-in for neuro-spicy people? For one, me and you are here. Like, hello... |
This dude, \[unintelligible 01:14:42.08\] stories about Steve Jobs... It's so funny, from hyperfixation food, to just like wearing one thing over and over again, because your brain doesn't have to do the work of thinking about what you're going to wear... And then you think about how many people in tech just wear a T-s... |
**Justin Garrison:** In 1937, Claude Shannon has his master thesis... And it is described as the most important master thesis ever written. |
**Autumn Nash:** Shut up... |
**Justin Garrison:** \[01:15:44.20\] In this master thesis je describes how you can take a switch and make Boolean logic circuits with switches. He basically invented -- this is before transistors, this is before they knew how to make AND gates, and OR gates... He described it. And he said "Oh, you can solve logic puzz... |
**Autumn Nash:** This is why I think real revolutionary moments in tech or technology or industrial or whatever happened in simplicity. That is timeless. Look at -- between him, \[unintelligible 01:16:23.09\] That was a time when we didn't have anything close to computers. But they were already thinking in a way that i... |
**Justin Garrison:** Yeah. And that was -- for Claude, it was about 11 years later. He had a paper called "A Mathematical Theory of Communication." And that is literally where he lays out what checksums are. |
**Autumn Nash:** That's amazing. |
**Justin Garrison:** How would you communicate two things over a wire? And he's like "Mathematically, this is how we can prove how this would work." And it was known as the Magna Carta of the information age. The very beginning of "We can send data to someone else and verify those things." And this is also -- he was wo... |
**Autumn Nash:** Which is wild, because so many ciphers have been -- and just the fact that encryption libraries have had to advance so much to keep up with it... Nowadays people don't even really want you to write your own encryption, because it's so easy to break them. |
**Justin Garrison:** Right. And so he was proving that -- again, this is without computers. Mathematically, how could we essentially encrypt something? How do we add noise to structure, and then send it to someone, and then make sure that they can get the same message out? |
There's a really cool -- if you go look up that cipher, the Vernam Cipher, there's a Python script that's like not even 50 lines. And you can actually follow it, like "Okay, this is where this changes, and this is how this changes..." I was like "That's fascinating", because it's so simple, but he proved it with the ma... |
**Autumn Nash:** The best stuff is simple, though. The best stuff in fashion, and design, technology - it's always simple, because it's really got the value there, and it doesn't need a bunch of other stuff to prop it up. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.