684: It’s Not What Young People Do  🔗 **Source:** [684: It’s Not What Young People Do](https://share.snipd.com/episode/d3c042ec-3803-4489-ac98-3f9afb124f1a) ## Highlights - **Running Tumblr On Dedicated Servers Before EC2** - Marco explains Tumblr's early infrastructure used leased dedicated (bare metal) servers rather than cloud instances. - From 2006–2010 Tumblr ran a few dedicated servers, never seeing a datacenter because hosting providers handled physical hardware. Transcript: Casey Liss First of all, how is it, Marco, that you had never seen a data center before? You worked for Tumblr and were kind of their infrastructure or IT person for years. What gives, man? Marco Arment Yeah, so this was a question posed on the Overcast Reddit. You know, to people who don't, who kind of aren't in this world or maybe are in a different part of this very large world, it might have seemed ridiculous that I would have operated Tumblr For its first four years and never had been in a data center before. And the reason why is because during those four years, 2006 to 2010, Tumblr didn't have its own data center. What we did was this was very early in AWS. EC2 didn't exist for at least the very beginning of that. But what we had instead, you know, what people did back before compute instances, which used to be called VPSs or virtual private servers, what we all did before VPSs was dedicated servers. So what this is, is now in the business, I believe they now call this bare metal servers. You go to a hosting company on the internet and you just lease a server, you know, through their web interface or their salespeople or whatever. And you have a server allocated to you that you never see. You never deal with the physicality of it. They deal with it. But, you know, and you have to do things like, you know, put in like a RAID controller and make sure you have multiple disks. And if one fails, either their monitoring or your monitoring, you'll have to tell somebody, hey, go replace this disk. And, you know, like you have to deal with stuff like that. Your server can break and you just, you're out of the stuff. If you don't have it backed up, too bad. There is not much of a concept of like upgrading to a different server because like well you can you know you can copy the data off yourself and copy it onto a new server but you know the the Concepts we have now like instances that it can be resized like that didn't exist you know very little concept of like managed images and stuff like that but this is how servers were run By many people at that time at this scale. At that time, it wouldn't have made any sense for us to buy, like to have our own data centers or our own racks and data centers. We started off the first few years of Tumblr. We're on rough. Most of the time we were on, we were on two servers, you know, maybe up to six or eight at, you know, by the end of the second year or something like that. I put the timeline out all in there. That's what they were. They were dedicated servers. And that's what everyone did before, you know, AWS EC2 and before compute instances, virtual instances, things that we use now. And again, that world still does exist. It's called bare metal servers. ([Time 0:04:06](https://share.snipd.com/snip/e0b2bd8a-dafd-44e0-845b-f11c561602a2)) - **Producer Transcripts Are Limited By Ads And Format** - Podcast 2.0 transcript tag supports WebVTT/SRT so apps can display producer-supplied captions. - Marco will parse those files but warns dynamic ad insertion and lack of per-word timing limit usefulness. Transcript: Marco Arment Yeah, yeah. The podcast 2.0 spec people defined a transcript tag, like a podcast colon transcript tag years ago. It can point to, for an episode, it can point to a SRT or WebVTT file. And when Apple Podcasts launched their transcripts, you know, whatever that was a year and a half, two years ago, they said, okay, well, we will do our own transcripts. But if you want to supply your own, we will obey this tag. If you use this tag in your feed, the podcast transcript tag, we will show that I think instead, I think not in addition to, but I haven't actually tested it. And the reason I haven't actually tested it is that very, very few podcasts so far actually do this not none but not a lot so i didn't feel that it was necessary for my first beta release I am in the process of building in the the architecture to support that so i do plan to show transcripts from people's podcast transcript tags from SRT and WebVTT files if they are present. I haven't quite decided, though. So one of the problems is if you include your own transcript, first of all, you then cannot use dynamic ad insertion because your timestamps will be all messed up. And there is not a well-supported standard to embed a transcript in a file that any podcast app would include. Now, there actually is an old MP3 ID3 tag, I think USLT off the top of my head, but there actually is, or no, that's unsynced, I think SLT, whatever it is. One of those tags is an ID3 tag to specify time-synced captions to be displayed. So that feature exists in the MP3 ID3 spec, but I don't know of anything that uses it. So if you were to support like podcaster supplied transcripts with dynamic ad insertion, something like that, that comes with the file and therefore can be adjusted with the file whenever They insert and remove ads, that would be the only way to really do that. But given that these platforms don't even like the DAI platforms don't even support chapters, which work the same way, I'm not holding my breath on that. I think that's very unlikely for any of them to ever do. So because DAI platforms generally won't or can't do that, and because almost all podcasts of medium to large size now are monetized with DAI, I can't imagine that podcaster supplied Transcripts are ever going to be that big of a thing. Now, this could be wrong. And another thing they could do is they could supply a transcript with timestamps that just don't include any ads, but then the real-time display of it will break in the player. Because if their transcript is just their content and doesn't have any of the ad content in it, which it wouldn't in this context, as soon as you reach a point in the show where they've Inserted an ad for, say, two minutes, from that point forward, all of the timestamps in their supplied transcript will be off by two minutes. And so the utility of that, I think, is always going to be significantly less than one that was dynamically generated in such a way that it can synchronize itself back to the ad inserted Version of the podcast that you downloaded. So in that way, I think my transcripts are probably going to provide more functionality than many supplied transcripts will. Because basically, yeah, if you supply a transcript and you have dynamic ads inserted, that's not going to play well together. So even if you supply your own, I haven't fully made this decision yet, but I think even if you supply your own, I think I should probably also offer mine and maybe just have like, you know, Like a tab picker at the top of the screen that says, you know, my transcript, their transcript, something like that. ([Time 0:07:07](https://share.snipd.com/snip/3cc86303-e3fb-4f62-ab44-fc5678ed55c3)) - **Prefer Proven Lightweight Tools Over Costly Managed Services** - Use simple, proven tooling for large-scale systems to avoid vendor lock and high costs. - Marco runs Overcast on MySQL, Memcached, Redis and Beanstalkd to minimize expense and long-term risk versus paid managed services. Transcript: Marco Arment So what does that mean? It means things like MySQL, It means Memcached, Redis, and Beanstalk D. That's my server stack. It's those things, you know, Nginx with the PHP FPM plugin, like that's what I'm using here. So what this queue is, is Beanstalk D, which is very old, very boring, and it works fantastically it has almost no features almost no one that i've ever met uses it but it's fine and i've Been running it forever i do use redis's cues for feed crawling because they allow me to the way overcast is feed crawling basically i have these crawl servers that run the go process That i wrote a million years ago and then instantly forgot Go. I learned Go just to write that one process. Haven't touched it since and have since completely forgotten Go. But, I mean, I think today if I ever needed to add anything to it, I'd just have AI help me because I need to relearn this entire language. But anyway, the way those work is they crawl the feeds, they stuff the content of a changed feed into Redis, and then they add to a big Redis queue so that other processing PHP processes That parse those can go pick them up and crawl them. Redis, though, is even heavier duty for this case than Beanstalk, which is Beanstalk d is like what if memcached d but for queues beanstalk is where i put lighter queue work in overcast That that is not feed crawls and that's things like checking for redirects uh it's things like checking for artwork updates on on feeds uh things like that um sending notifications Sending processing my ping API, stuff like that. And so I built transcripts on top of the Beanstalk section of Overcast. Each of the Q consumers that's running on the Mac minis just hits an API endpoint on Overcast web service and says, hey, give me some jobs, give me some jobs, give me some jobs. Fairly straightforward setup, I think, for Q processing. Now, again, there's lots of other queue services, queue stacks, things like that. I haven't used the services because they're just too expensive. Again, what Overcast processes overall in its job queues, especially if I include feed crawling, we're talking over a million jobs a day, easily, any kind of like managed queue service, When you actually price out what it would cost, it's a lot for at this scale. So I don't I don't use those, I just run it myself and it costs nothing because running Beanstalk D is incredibly lightweight. It costs nothing. Same. That's the same reason I send Apple push notifications myself. I wrote a simple PHP class to do it. It's not that hard. And, you know, I managed it through Beanstalk D and I send them and it's fine because I send probably millions of push notifications a day and it costs me nothing because like the the processing Time the server resources to send those is as close to zero as you can imagine like it's just nothing whereas if you can go to some service to do it for you and you start paying oh it's you Know just x cents per thousand notifications you know whatever their pricing ends up being for most apps that's fine when you're doing millions of things a day that could be like tens To hundreds of dollars a day. And when you spread that over the 12 years I've been running Overcast, that's real money. So if I can write it myself in less than a day and have it work entirely with my own stuff, then it costs nothing to run over time. I don't have to worry about like what happens if this service gets merged with some other company or shuts down or sunsets or, you know, changes their pricing model. And now, now they are, now it's into enterprise and costs 10 times more for my use case. Like all those things that happen constantly in our business, I don't have to worry about them. So that's, that's why I do a lot of this stuff, like the boring old way with a few simple processes running on some Linux servers that don't need that much from me. It is not cool. It is not sexy. It is not trendy. It is not what young people do. ([Time 0:13:52](https://share.snipd.com/snip/d2387354-7815-4674-9c0f-a3d4e08faada)) - **Custom Transcriber App Running On Each Mac Mini** - Marco built an Overcast Transcriber app that runs on each Mac mini to pull Beanstalk jobs and run Apple's Speech Transcriber. - Each worker checks in with the server, reports stats, self-monitors and restarts via launchd when needed. Transcript: Marco Arment Yeah. This is the actual thing that is running on the Mac minis. This is the app that transcribes things, that pulls the jobs from the servers, and actually transcribes them. Casey Liss Oh, interesting. I'm sorry. So this is the app that's actually doing the work. Oh, my mistake. No wonder we were confused. My apologies. Yeah. Marco Arment One copy runs on every Mac mini. I assign each one because the Apple Speaks Transcription API only supports three languages being installed at once on the system for reasons I don't know. Maybe that made sense on the iPhone. For some reason, it's also enforced on the Mac. But anyway, so it supports six languages, but only supports three of them being installed at once. So I have a simple picker on top to choose which three languages. They all do English, and then they all kind of split the other five languages because there's just there's way more English than any other ones. So that's what that is. And I pick how many jobs I want, and I run them. And whenever this if this app ever crashes or if the computer reboots no problem launch g just starts it back up again and uh it's totally fine and then this app checks in with my server With my main servers each one of these checks in like you know once a minute or something like that and it reports its stats of how many jobs it has done uh the last minute or whatever. So that way, how many minutes of podcast has it transcribed? That way the servers can then kind of watch out for anomalies. So if, for instance, one of the Mac minis doesn't respond at all, it will alert me and I can go reboot it or something. That doesn't happen very often. But also if one of the Mac minis is reporting like a suspiciously high or suspiciously low job count per minute or transcription minute rate per minute, I can then go take a look and see Something might be wrong there. And then the app itself, the Overcast Transcriber app that you see here, that app also tries to monitor itself. If, for instance, it can't get jobs for a long time, for more than a few minutes, or if a job never completes and it never refills its jobs after a certain amount of time, it will quit itself And let Longstay restart it. ([Time 0:18:52](https://share.snipd.com/snip/0c8b7f5a-2efe-4e99-8178-8f52727b41a4)) - **Distributed ML Needs Fast Interconnects Not Just More Macs** - Distributing large ML models across many Macs is feasible but network latency and bandwidth matter most. - Marco's minis are base models with 1Gb Ethernet and TB4, so Thunderbolt 5 RDMA (fast local links) works far better than Ethernet clustering. Transcript: John Siracusa Oh, that's just the old Slashdot thing. Marco remembers this, right? I heard about them i don't think i ever actually have seen one uh the meme was and i'm surprised there was not a like know your meme uh page for this i guess it's so old it's such an old meme That like it's not worth documenting as a meme on the internet but uh beowulf was uh like a we'll put a link to the wikipedia page it was a system where you just take a bunch of commodity servers And then distribute a job across them and you say you'd run linux on these random pcs and you'd be like but now i have five pcs and if i have a job and i can break it up into five pieces these Five pcs can work on it and so all the slash dot nerds back in the 90s were like oh linux is awesome and beowulf was awesome and anytime there was any kind of hardware story on slash dot uh I don't know when it started but people started posting imagine a beowulf cluster of these and that just became a meme where people would say it about things that are not computers or Whatever. It started half serious and became a meme. And so Aaron Dippner basically just wrote, imagine a Beowulf cluster of these about your Mac minis. You could take one job and distribute it to all 48 and they'd all work on it at the same time. And what he's referring to, we'll put a link to this in the show notes is uh this thing where apple gave a bunch of youtubers a stack of mac studios you've probably seen them in the thumbnails I think it was like four mac studios with like a little little miniature little cute little uh mac studio rack kind of um and it's to promote their rdma over thunderbolt technology that Was added in mac os 26.2. We'll link to Apple's developer site about that tech. What they say about it is RDMA over Thunderbolt enables low latency communication between Thunderbolt 5 hosts for use cases, including distributed AI inference using MOX. And so what the YouTubers did with this stack was they got like the biggest model that they could get that wouldn't run on a single Mac studio because it didn't have enough RAM and they Would run it on the four Mac studios combining their RAM and you know do performance numbers or whatever we'll put links in the show notes to two videos one from Alex Ziskind uh the title Of his video is I ran a trillion parameter AI on a Mac and then Jeff Geerling also has one that says Apple didn't have to go this hard but there were others many YouTubers got this so if you Want to see an example of a technology that Aaron is talking about imagine a Beowulf cluster of these and it's the same type of deal it's like can you distribute something across multiple Computers that wouldn't fit on one and can you do that quickly and obviously doing it over Thunderbolt 5 with machines that are right next to each other is really fast doing it over Ethernet Is probably a little bit slower, but that's what he's pitching here. Marco Arment Yeah, and so the main downside of this for my use case is that, yes, as John said, speed of the communication is everything. And, you know, Ethernet between these Mac minis is just one gigabit because these are the base models. So they don't have 10 gigabit ports. I also don't have a giant 10 gigabit switch there but you know that could be remedied um but they don't they only have one gigabit ports and these mac minis only have thunderbolt 4 because Thunderbolt 5 on the mac mini currently requires the m4 pro chip which like triples the price of the mac mini and does not triple the performance for my use case. So I didn't get them. So because I have the base models, I could only do it over Ethernet. And that would be probably slow enough that would probably not be worth doing. John Siracusa And the pitch for this is that you would somehow use some bigger model than what you're using, because the model you're using now runs on your Mac Minis. It fits in 16 gigs of RAM you know runs on the phone like the pitches well what if you wanted to run some more sophisticated model um that wouldn't fit in the ram of a single one you could you Could take lop off a portion of your 48 and say this is like this cluster of four is the the big beefy uh transcriber for the top 10 most popular podcasts and we run a bigger model there Yeah, Marco Arment Exactly. But right now, this is something that I'm going to look at these kind of uses in the future, especially because once I'm done transcribing the back catalog for as many podcasts as I can, These are going to be at a third to half capacity. So I'm going to have a lot of extra computing power. And I mentioned before, I would definitely want to move some of my heavier tasks from the main servers that are, you know, the main Linux servers. I'd love to move a lot of those tasks to these, you know, to whatever degree that's reasonably possible. But that's not going to work for everything. So I'm still going to need a bunch of servers. So I will have to maybe get creative on like, what do I do with this computing power that I have here? And, you know, ideally, like things that can stay within one machine are better because, you know, it's easier to scale that. It's easier to share code between the iPhone and the servers. You know, so ideally, I would just keep doing more things with Apple's foundation models. And we'll see like, you know, this summer at WBDC, there might be updates to the foundation models and maybe they get things like larger context windows, which should make them more Useful in a lot of, you know, ideas I have on how to use them. So we'll see. We'll see how these go in the future. Right now, they are being criminally underused in the sense that they let's see my current, my memory use, I just pulled one of them up here. One of them that's under like, you know, full transcription load right now is using about 7 gigs of its memory, of its 16 gigs. So I have about half the memory that I could play with. Also, they are really not using their GPUs much at all. That will change once I move image resizing to them, but it won't change that much. So I think I'm going to have a lot of CPU or a lot of GPU capacity that I could use and a decent amount of RAM. So we'll see how I can play with that over time. ([Time 0:21:39](https://share.snipd.com/snip/9a6cba02-1f78-43c0-a1f7-5df5eedd57bf)) - **Use Apple Enrollment And MDM For Zero Touch Mac Fleet Setup** - For fleet Mac management use Apple's device enrollment + MDM for zero-touch deployment and central control. - Marco learned you can register serials via Apple Business Manager and push OS/configuration remotely, avoiding per-machine setup. Transcript: Marco Arment So what companies do that have to manage large fleets of Macs, of course, is not setting them all up one by one like I did here. Like that's, no one does this. No one should do this. What companies do is they have this thing through what I believe until yesterday was called Apple Business Manager and or something like that. There's a lot of words that all sound very similar. But Apple has like a business API. You can configure machines to be set up for your business and locked to your business with MDM, which is the framework they use to manage like corporate iPhones and iOS devices and Macs. You can buy these things through your Apple business rep, through Apple's business channels or through certain authorized resellers and have the machine's serial numbers automatically Enroll to your company when they connect to the internet for the first time. So you can, they call it like zero touch deployment, I think, so that you can like ship a laptop to a remote employee. And when they plug it in and set it up, it gets activated to the company's management system through Apple. Casey Liss That is very cool. Marco Arment And gets all the stuff it needs to get right from there, which is very cool. And once you are in these systems, you have full management over them. So you can do things like push software updates and stuff like that. And you can run programs, you can change settings, whatever it is. You can have all that centrally managed. So that is basically what I want. And the only downside is when you already have the machines pre-existing, like I do here, everyone said like, in order to enroll them into this kind of management, you need to wipe them And you need physical access. So it would be a process of, you know, I would like, you know, go, you know, go to the data center one day, bring a big cup of coffee, you know, keep it below the Mac minis, John, and just go Through like one by one all right reset this one reset this you know just take take them down one by one reset them wipe them put them put them in the thing um but the before this announcement Of apple business which happened you know yesterday the day before um yesterday before this announcement there was like the second part of it which was okay apple business manager Would have like the the component of like registering the serial numbers associating them to a business that had an mdm thing but then what everyone else did was have their own mdm providers Through companies like jamf former sponsor of the show a long time ago um this is what most big businesses do is they have one of these mdm apps that the problem with those from my use case And, you know, they provide a huge amount of functionality because usually it's not only providing like the basics of MDM configuration and stuff like that, but they also usually have Additional services they layer on top of it, like security packages, stuff like that. The problem is that all of those MDMs, or at least most of them, you're paying something like six to $12 per Mac per month. And, you know, again, you have 48 Mac minis that literally don't have users. That's not a great selling proposition for me. Like all of the features of these MDMs that do like, you know, user security management, making sure your users have good passwords and making sure they don't browse weird things on The internet. Like that's all stuff that businesses do need for their computers, but I don't need for these computers. These computers have no user. They run one app. You've seen it now. Like It's a very simple arrangement. And what I'm doing right now to keep them up to date, their software I don't really keep up to date yet because that hasn't really had a need yet. I guess I'll get there. And that's why I'm looking at something like an MDM, mainly for OS updates. ([Time 0:31:02](https://share.snipd.com/snip/ea8b59de-22ba-49d7-848f-0283c74f26dd)) - **Automate Fleet Tasks With SSH Scripts Or Ansible Before Paying MDM Fees** - Before paying per-device MDM fees, try lightweight automation tools for your specific tasks. - Marco uses SSH+scp+launchd scripts and Casey suggests using Ansible/other Unix tooling to run commands in parallel and verify results. Transcript: John Siracusa One suggestion, and a lot of other people wrote in with a suggestion as well, is it does seem like you need something probably from Apple that can handle OS updates and stuff like that. Although the various reports we got about the quality of Apple's previous Apple Business manager software and their mdm stuff is not all glowing but you know and this is this is a new Product apple business supposedly so there might be some growing pains there but the other part of it the part where the part where you're scp-ing a file onto a bunch of uh macs like everyone Uh this is where the second piece of advice we got is like unix nerds are saying there's a million you know free open source command line packages for uh configuring a whole bunch of unix Servers and yes they're not you know it's a mac but it's also a unix server and if all you have to do with if all you're doing is essentially something that you can do with scp now rather than Doing scp in a loop and dealing with that and everything there's a million software packages most of which probably run on the mac that you can try a lot of people suggested ansible there's Also cf engine back in the day there was puppet i don't know if it still exists a lot of those things do have a way to run on the mac they're you know they're unix command line things and you Have to figure out how to get that installed on the Macs. But if the only job of that part of the system is, oh, and by the way, there's something that I want to do that I can do from like a shell script or whatever. Just you can do that outside of Apple's things. And the reason I worry about it, having not used these products is like you have a new version of the transcription app that we just talked about before. You want to get that to all the machines. I'm sure Apple Business Manager has a way to get the new version of your transcription app to all the machines. I'm also guessing that it takes its time because in a business environment, everyone needs the new version of Outlook. It rolls out when it rolls out. There's no like I want it to go to all 48 machines now now now yeah i can't imagine it's taking 30 seconds i i yeah even if they offer that as an option it's just like that's not what those things Are made for like you you probably never even want that in a situation where you're deploying to 2000 max you don't want them to all instantly get the new version of outlook you want it To roll it out so maybe depending on if apple business manager or whatever if apple business has some gaps rather than saying, Oh, now I need one of those commercial MDMs that charges Me an amount per Mac per month. At least look at like Ansible and some of the other Unixy stuff that basically are a fancy distributed way for you to run Unix commands on a fleet of servers without you having to like deal With the plumbing. I mean, so my argument so that would be though like i already built that with ssh and a shell script like it's you know but like it's just you know how good is your error checking and how how Paralyzed is it how fast can it do all 48 machines yada yada like there's a reason these things exist people things like puppet sprang into existence because people used to write scripts That would loop over things and distribute tar balls and untar them. And then they made products out of that because doing it manually kind of sucks. Yeah, I mean, and again, this is the kind of thing, what your needs are determines a lot about what the solution here should be. Marco Arment Needs are so simple here that if all it's going to save me doing is like you know right now like i i have just a simple you know text file that's about 30 lines long that is my instructions For setting up a new mac mini and there's not much it's not that hard you know it's like i said i changed like five settings and you know mostly about power management like stay on really Stay on no really always stay on and and you know give it my ssh public key and say all right well you know accept anything from this go and oh and turn on remote desktop that's about it like That's roughly all i do if something is like well you know you can you can more control over it by installing this different product. Okay. But right now I install no different products and I have all the control I need. So to me, I would say like that additional complexity, even if the product is free, you know, setting aside any kind of cost, the additional complexity of having another thing being Involved, that's another moving part. That's another company that can go out of business or another thing I have to keep updated or that might break in three years with new OS update that Apple gives or whatever. Every one of those moving pieces I evaluate, like, is this giving me enough benefit over my current situation to be worth its additional risks and complexity? And because my needs for these are so simple, the answer to that for a lot of these things is no. Now, not for everything. And again, software updates are the big thing that I definitely would like to be automated better. But adding any other layers, any other packages, any other products, they all have to pass that test before they're going to be worth it for me. And they're going to be worth it for a lot of businesses before they're going to be worth it for me. John Siracusa Yeah, I wasn't suggesting you get rid of your little 30 line thing that you run. I was suggesting you just get something else to be the plumbing to run your 30 line thing. That's it. ([Time 0:38:15](https://share.snipd.com/snip/c75a8872-4c5e-4e74-8822-3731aef3d82f)) - **ShazamKit Could Replace Custom Audio Fingerprinting** - Apple offers ShazamKit which provides audio fingerprinting and alignment similar to custom solutions. - Marco plans to experiment because his fingerprinting needs quarter-second granularity and small signatures for long shows. Transcript: Marco Arment All the you know bugs and reports and the other thing is i said uh that you know that i was i i had spent a huge amount of time last summer getting an algorithm to fingerprint the audio so I Can align it around dynamic ad insertion. I did not realize that ShazamKit was an API and it offers exactly that functionality. Oh, no. So I haven't actually, like, I am going to try it. I'm going to see, like, you know, is their signature better than mine? Like, is their fingerprinting? Is it better? I don't actually know. So I'm going to experiment. I'm not going to do that, like, this week or anything. But sometime soon I'm going to experiment with that to see, like, should I basically start using their alignment instead or in addition to mine? But that was kind of a little bit like, oh no, did I spend months doing something from scratch that Apple has an API to do? We'll see. Casey Liss Yeah, I got it. I know Guy Rambeau had reached out to us to tell us about Shazam Kid. Perhaps you heard it from others as well. Marco Arment Yes, that's right. It was from Guy. Yeah, that is indeed where I heard about it. And yeah, I'm like, oh, no. But we'll see. So part of my fingerprinting algorithm is it has to be very low granularity of time. If the block size it's working on is a 10 or 15 second chunk, that might not be good enough for what I'm doing here. Like what my fingerprinting, the sample resolution for it is something like a quarter of a second. And it also has to be a small enough data set that like whatever the fingerprint that you generate for an entire podcast, it has to be a reasonable size. It can't be like many megabytes of data if you're going to do that resolution across a three hour show. You know, so I'll see, you know, I don't know what what exactly their data is stored and what resolution it is. So that's all going to be part of the experimentation phase and see, like, you know, should I switch to there or not? ([Time 0:44:20](https://share.snipd.com/snip/222709fc-744b-4033-9284-38b991268a5c)) - **Many NYC AirPods Max Appear To Be Street Knockoffs** - Casey and listeners noticed many AirPods Max seen in NYC are likely knockoffs sold on street stalls and in small electronics shops. - Marco says at distance it's hard to visually tell fakes from genuine units. Transcript: Marco Arment I, so many people have written in to say either they, either they suspect their fakes or like they live in Manhattan and they see that they're fakes. And I can, I can verify when I'm in Manhattan about, about once a week. Uh, I do see like the, you know, the guys in the streets that sell all the, like the fake luxury goods on those little tables yeah there's always airpods max lookalikes on those tables Too and every like all like the store windows that are like the like the random little electronic stores there you look and you like i bet most of the stuff in there is probably knockoffs Like you know they also all have airpods max lookalike headphones so i'm sure many air Max looking fakes are sold. But I'll tell you, I'll be damned if I can spot one. Like people say like, oh, you can tell they look cheaper, their finish is wrong, they have a plastic crown, like different things people have told us to like try to spot them. But at the distances that, you know, I'm not going to like, you know, hover over someone's shoulder and start eyeing the details of their digital crown to see like, Hey, is that plastic Or metal? Can I tap that for a second? You know, like I'm not going to, you know, at, at regular distances, I can't tell. ([Time 0:46:50](https://share.snipd.com/snip/5790b4f4-0785-418b-a762-a4c5161d1046)) - **Use Passkeys And Store Backup Codes Offsite** - Adopt passkeys where available and keep backup recovery options like backup codes or a memorized vault password. - John says passkeys remove server-side secrets, reduce phishing, and sites usually provide one-time backup codes to store offsite. Transcript: Casey Liss We haven't had the opportunity to do that in a little while. And we're starting with something that could have been written by Casey List, but was actually written by Philip Miller, who writes, Paskies have me befuddled. Amen, brother. When they first emerged, I took a wait and see approach. And that's where I remain today. Again, amen, brother. Even though many sites and services pressure me to adopt Paskies. Are all three of you fully embracing Paskies now? What are the biggest pitfalls of, or excuse me, what are the biggest benefits of passkeys for someone like me, who's a faithful user of tools like 1Password for managing traditional Passwords and two-factor codes? What are the real risks of passkeys? Am I being foolish to worry about scenarios where my Apple devices are all lost in a house fire, now I'm locked out of every account? With 1Password, everything is backed up in the cloud, and I can keep my emergency logging credentials on a single sheet of paper in a safe deposit box. Are passkeys equally resilient? And I will also add a fourth question. What about family passwords? My understanding is that you can share passkeys or maybe not originally, but certainly now. But there's a lot of stuff that, you know, one of the great things about 1Password is that it has family vaults or you can get a family account with a family vault where, know, Aaron and I will share a bunch, if not most of our passwords. So I know nothing about this, even though I should. I don't know, Marco, if you have thoughts on this, and let's start with you if so, but I'd very much like to hear John's perspective as well. Marco Arment I have some thoughts, and then we'll kick it to John, who will have the right answers. The way I've embraced passkeys is when a website asks me, hey, you want to make a passkey? It's much better, blah, blah, blah. I will usually at that point say yes and do my fingerprint or whatever. I have also found that as I am still bouncing between Apple password or Apple passwords and one password, although more one password these days since I'm still using Chrome as my default Desktop browser, the way both of them handle pass keys is fine. You know, with Apple, you go to a website, if it wants a pass key, it'll show you the standard like touch ID, face ID prompt. And you, you know, you do you authenticate that way and it gets you in with one password. If you have the browser extension, it just if it's unlocked already, it'll just show a little thing in the upper right corner and it's like, do you want to use the passkey? Click to log in. You click to log in and it logs you in. Both of those are fine. The 1Password one is I think a little more convenient of just providing the security in a different way but they're both fine. My main issue with passkeys is that it's kind of like that old XKCD comic that's like, we have too many standards. Here, I made a new standard. From a technical and security perspective, they're amazing. They're way better than everything we had before. But from the actual user experience of using them in practice most of the time, it's just another thing that gets added to the pile of things that we are supposed to keep track of and, you Know, having a password manager or something and, you know, try to share maybe with mixed success with family members or coworkers or whatever. It can be done very well in an ideal implementation of passkeys, both on like, you know, the platform side and on like the website side, they're better than everything else. The main downside with passkeys so far is that the way that I see them being implemented by websites and services is not as a replacement for passwords, but as a replacement for two-factor. So it's actually just taking this other complex thing and replacing it with a different complex thing. Now, the security of a passkey is way better than most of those other systems or alternatives to it. But the user experience of it is still really clunky in a lot of places. And ideally, would replace the entire login like ideally you could just have it like automatically if you have a passkey for the site you're on and it shows you a login screen it just prompts You to use it and it logs you in all the way that would be the ideal case and you occasionally see sites like that but it's not it's not the common case um so that's that's of my kind of mixed Opinion of them is that they are a great technology that solves a lot of problems. They also create a lot of inconvenience in certain contexts and they are not really making logins on websites that much easier in practice, not because of their inherent technology, Because of the implementation decisions of those websites. John Siracusa Yeah, I've really come around on past keys for the past six months or so. I was always trying them out whenever they were offered or whatever, but now having implemented them or having asked coding agents to implement them for me more precisely and seeing All the details there and and all the different trade-offs and everything um i still think implementing them is the most annoying part of pass keys because the input like if they want To get better adoption on websites and stuff they should really you know streamline that and work harder to give like friendly libraries that are easy to use with good documentation Casey um but anyway the coding agents can do it so it's not that bad um but just a quicker view like what's good about pass keys why would you want to use them instead of passwords or whatever They have a bunch of security advantages that uh passwords and two factor stuff are never going to be able to match just because they're inherent in the technology the first is that um When you use a password on a website, the server stores something on its side that could potentially compromise you. You hope they don't just store your password in plain text, but as we've seen, some websites do that. And you have no way of knowing if the website is incompetent in storing your passwords in plain text. So there's that. Second, so they don't store it in plain text. Let's say install a store encrypted hash of your password or they should be doing what algorithm are they using for it has the algorithm been broken was the algorithm secure 10 years Ago but now it's not anymore and all this is relevant because websites get hacked and what people leak from them is account information which is usually your email address which whatever People don't like it to leak but you know your email address is not a secret uh but also some hashed thing for your password and you're like well that's secure they can't get my password From that can they well again what algorithm are they using how maybe they can't get it now but three years from now they'll be able to do it with some nvidia gpu or something that can crack It pass keys do not store secret information on the server again you're assuming that they're implemented in the same way but i can tell you that pass keys don't involve the server storing Something that is secret the server what the servers store it's public information they store your public key it is anyone can have it it's not a secret you can give it to the world and They use that to verify through this protocol that you're being signed in or whatever um so if a website gets hacked the only thing they can leak about you is your email address which again I don't think is actually secret because spammers just spam every email address in the entire world they just go through every combination of letters numbers like anyway um ([Time 1:10:12](https://share.snipd.com/snip/88683eb8-35f6-4e06-8f32-e752bfe15a45)) - **Third-Party Apps Drive Mac RAM Usage More Than macOS** - macOS memory pressure is often caused by third-party apps, not the OS; Electron/Chrome apps dominate RAM use. - Marco notes Apple can optimize OS swap and VM, but developers of heavy apps rarely prioritize low-memory efficiency. Transcript: Marco Arment Yeah, I think the main thing to keep in mind here is that when you look at where RAM goes on a Mac, macOS is not what you have to worry about. What you have to worry about is all the third-party apps, especially apps that use things like the web frameworks, like Electron and stuff like that. So you have Chrome itself. Chrome is a big resource hog compared to Safari in most ways. And then all of the Electron-based apps, they just gobble up RAM like crazy. There's not a lot Apple can do about that, really. The OS does its best, but for the most part, that's what you have to worry about. And now, that being said, when you're designing an app like that, you look at the installed base of what people actually use. And if computers that many people use are going to run that kind of app very slowly and it's going to bloat their computers up um you don't care because the entire industry doesn't care And you just have to suck it up as the user you like you know you want to think that they would look at you know the entire field of computers out there and be like should we maybe not use all The ram in the world should we maybe put a second of effort into not duplicating this thing over here? Nope. Nope. No one thinks that way at all. No one pays attention. Not a single software developer ever cares about that kind of stuff when making things like, you know, various electronic apps and things like that. So, like, Apple does care a lot. No one else does. Yes, the 8 gig RAM constraint in the MacBook Neo keeps macOS lean, but macOS wasn't your problem. Your problem is everyone else and they're going to do whatever they're going to do. And the best thing macOS can do is have really good swap and virtual memory support. ([Time 1:25:56](https://share.snipd.com/snip/876b9c56-d5f8-432e-a2c5-1baa0d73e7df)) - **Dual Mode Phone Mac Would Be Confusing Notably Hard To Sell** - A phone or iPad that switches to macOS when docked sounds appealing but Apple likely avoids hybrid devices for UX and product clarity reasons. - John argues different OS modes cause confusing UX, storage and synchronization complications compared to cloud-first sync approaches. Transcript: Casey Liss Finally, Keith Heaton writes, how long until we see an iPhone with an M chip or an iPad Pro with the ability to plug into a Thunderbolt dock or a studio display and present Mac OS to you? I just know that a Mac inside your pocket will eventually happen. Michael Cook writes, does the MacBook Neo now mean that Apple will likely never allow the iPhone to plug into a monitor and be used like a computer the way you can with an iPad or a Samsung Dex? I just, I think this would be awesome, except I don't think it would be as awesome as I think it would. And I really just don't see Apple doing it. I think having this mythical like dual boot or dual mode iPad be incredible, but I just really don't see Apple doing it. I mean, they're steadfastly refusing, mostly, to allow windowing on iPad OS. I know obviously that's different now, but they certainly steadfastly refused it for years. I just don't see this happening. And I don't think it really has anything to do with map canio i just don't think this is apple style it's like incoherent from a software message like technically possible sure but it's John Siracusa Like apple seems not to and i kind of agree with them make um toaster fridges jekyll and hyde kind of uh devices where like it runs two different os's yeah sure you could do that from a technical Perspective but it's weird that like it's a phone when it's in your hand but it's a mac when it's not and like how to synchronize the world of the phone and the mac with each other and obviously You're taking up more disk space and you kind of need to have power being supplied when you're hooked up to the monitor too because otherwise it'll drain your phone battery and like it's Just it's a weird product to pitch even though this is exactly what i pitched in one of my early mac world back page articles which was like a laptop with a screen that folds back on itself And when you fold it back it runs ipad os because those those are the pieces available at the time and it's an interesting idea but like apple never took us up on it i think what people mostly Are looking for here is like either they just want like an ultra portable like mac like experience that goes everywhere with them and like maybe the folding phone will help with that Or whatever or they're just really what they just want is like wherever i go i want my whole setup with me and they're just what they're asking for i think is what apple would hopefully Deliver well i don't know over the years is a better cloud sync experience like because that's that's the way this is done like in the google world for example wherever you go with your Google account there is in if you're tied into the google world there is all your stuff you got your mail you got your google photos you got your google drive you got your google docs like It doesn't really matter where you land all your stuff is there whether it's a phone or tablet or whatever and and if you have that experience and you know if that's working well you don't Need to plug your phone into a monitor. All you need is to be able to just like log into a guest account on that monitor thing and just get all your stuff and use it kind of like more like a thin client or whatever. But I just think this is a product choice that Apple doesn't want to make. And like I said, I kind of agree with them because it's just weird to run ios and mac os off the same device setting aside the the amount of disk space you're wasting and having to divide It up or whatever it's just it's not an efficient use of resources and it's a difficult product to explain to people and it's difficult product for people to like if you had this and you Used it you'd be like yeah but it is kind of weird that i run mac os and then i run the iphone os and the only place they sync is through the sipping straw of like iCloud stuff. And it's just I don't think it's a good idea. I don't think it's going to happen. Marco Arment Yeah, I think that's the kind of product that it sounds like a fun, interesting idea. But I think the reality of it would be weird and would disappoint people. ([Time 1:31:57](https://share.snipd.com/snip/aed7565d-c848-4f47-8f26-461076c1a31c)) - **Ensure Offsite Backups For Home Lab VM And Container Data** - Back up home lab VMs and containers to a separate system with redundancy and offsite options. - Casey uses Proxmox Backup Server saving to his Synology for now, but acknowledges cloud or offsite backups are crucial for disaster recovery. Transcript: John Siracusa Might have to get a new sound for you because speaking of spinning disks in your Synology, it made me think when you were rattling off all the things that you've consolidated onto this NUC. How are you backing this NUC up? Casey Liss It's funny you bring that up. There is, what is it called? Proxmox Backup Server, PBS, I believe, where it has its own mechanism for doing this. And basically all of that is going onto the Synology. John Siracusa So, I mean, the reason I ask is because you have so much set up, like the time you'd invested into setting up those 20 containers and consolidating everything. And it's not redundant storage, like the Synology spinning disks, the old cruddy rust that presumably you're not using RAID zero on the whole thing, I hope. So you have some hardware redundancy. Oh, what if the SSD goes bad? What if a kid spills a drink on it or whatever? Do you want to reset up your entire life on that NUC or do you want to restore from a backup? And then the question is where you put the backup. And as you noted, it is on the Synology. So you may not be out of the woods yet. Casey Liss No, but that is actually one of the only things that the Synology is still doing other than drive that would be hard to replace. I would probably have to do something else because... John Siracusa You can do a cloud backup or whatever. It's just a question of making sure it's some automated process. You actually are protected because one of the things that synologies have going for them is that if a drive fails, you're protected. Two drives fail, maybe not. Casey Liss Yeah, no, you're exactly right. I have no reasonable way to contest what you're saying. And that thought has crossed my mind. ([Time 1:50:35](https://share.snipd.com/snip/7faec546-1393-4a06-8ee5-7f0f91042dd4))