0:00

We've. Both been traveling and I've never

0:02

been quite happy with self hosted

0:04

tools around. Trip. Planning

0:06

maps, Or. Anything that

0:08

really is just better. Than.

0:10

A good paper map Alex You know it's just

0:12

sometimes when you're traveling special on the road trip.

0:15

It's. Really hard to beat the paper map or

0:17

hard to beat google maps. Les mis, just be

0:19

honest about it. That's. The other problem. And.

0:22

I don't know if I'm there yet, but I have

0:24

been playing around with something called for silly map. And.

0:27

Open a link to this in the show. Notes: It's

0:30

what they say. Privacy friendly, open

0:32

source, versatile online map. And.

0:35

What it does is a polls in a bunch

0:37

of different services that are based on Openstreetmap and

0:39

allows you to find things and visualize thing in

0:41

in ways that Google map. Never.

0:44

Does and there's a couple of terrain options in

0:46

here. The. Just make it.

0:49

While. It's it's you got you to go look

0:51

at why at Washington State and a couple of

0:53

these different terrain modes. It really helps you understand

0:55

like where the mountains around where the peak elevations

0:57

a rat but also you can and set way

0:59

points. You can bring in Gp X files, And.

1:01

Then lay them over this you can course and recall

1:03

those things. later on you can run it. Through.

1:06

Our simple docker compose to so it's pretty easy to

1:08

get up and going. Gets a little more complicated if

1:10

you want to put a pretty big database behind it,

1:12

but. Small. Instance, it's not

1:14

too bad. I always forget how good Openstreetmap

1:16

is. I just pulled it up. Now us

1:18

we're recording and there's a whole bunch of

1:20

information I am. I'm pretty familiar with the

1:22

air around even. Have been here five or

1:24

six years now. I know what the area

1:26

around me looks like on Google maps. An

1:28

Apple Maps been particular. But.

1:31

You know this. This is just drop me

1:33

off where my I S P spits me

1:35

out my browser as opposed to we were

1:38

actually live. I'm I'm looking at this openstreetmap

1:40

by overlay in the and set facile map.org

1:42

the bailing can the show notes I just

1:44

love the way that. Openstreetmap presents

1:47

it's data. It's not quite as

1:49

good as a good Ordnance survey

1:51

map. I have a bit of

1:53

a upon shown for good Ordnance

1:55

survey map, but you know this

1:57

is a pretty good digital alternative.

2:00

One that I thought was kind of interesting is to

2:02

go look like at a destination and then turn on

2:04

the hiking trails overlay. They have biking

2:06

trails as well. Oh, there's a good spot to hike. And

2:09

I looked and there's some massive great hiking spots near

2:11

me. So I forgot all about that. It's

2:13

neat and it has all kinds of different filters

2:15

and features. And you can host it yourself. That

2:18

and also a shout out to GPX pod

2:21

and GPX edit, which are two different apps for

2:23

your next cloud instance. And these

2:25

let you say import your Garmin nav. Maybe

2:27

you've saved favorite spots and then you can

2:29

view them inside next cloud. So there's a

2:31

couple of ways to do this. And I'd also

2:33

be curious how anyone out there listening has solved

2:35

their own self hosted map. POI,

2:38

you know, save the spot

2:40

for the future type tool. For many

2:42

years, my solution to that problem has

2:44

been drop a star on Google Maps.

2:46

I would love to get that brought

2:48

in house and own that particular data

2:51

mine. Yeah, yeah, that's what I'm

2:53

thinking. It's got to be a way. Get

2:57

grist.com self hosted grist is the

2:59

open source alternative to air table

3:01

and Google sheets. No shame here.

3:03

But if your company has one

3:06

of those load bearing spreadsheets, I

3:08

swear just about every company has one of these.

3:10

They become impossible to maintain over time, but they're

3:13

too important to fail. Everyone seems

3:15

to rely on one of these. And inevitably,

3:17

it's like one person owns it. And then what

3:20

happens if they go or like what happens if

3:22

they're bad data hygiene practices make its way into

3:24

the spreadsheet? What you really

3:26

want is something that is

3:28

easy to use as a spreadsheet

3:30

by all users, but as reliable,

3:32

accessible, maybe as portable as an

3:34

actual database. And it makes

3:37

sense that people get trapped in the spreadsheet

3:39

world. I've seen it. I understand. You got

3:41

to check out grist. It's really good at

3:43

combining why people like spreadsheets with why databases

3:45

make sense. And the user doesn't need to

3:47

know that it's a relational database underneath all

3:49

of it. They just get to build. And

3:52

if you've ever looked at something like a database and

3:54

thought, Hmm, I wish others could access this with a

3:56

GUI and then maybe not have to learn SQL or

3:58

something like that. Grist fits

4:00

in really well. Grist is

4:02

a fully integrated spreadsheet UI built on top of

4:04

a portal SQLite file. Non-technical users

4:07

can add and analyze data like a

4:09

spreadsheet, but everything is relational

4:11

and managed within a sophisticated set of

4:13

access rules, even down to

4:15

the table. So Grist works so well

4:17

with your existing authentication infrastructure, and it's

4:19

easy to integrate because of REST API

4:22

and there's lots of popular integrations ready

4:24

to go right now. And as

4:26

a listener of this show, I bet you're gonna

4:28

appreciate that Grist is open source and receives active

4:30

code contributions from the community like the French government,

4:32

who uses Grist for thousands of their employees. Grist

4:35

is the open source alternative you

4:37

can host yourself with a portable

4:39

file that makes sense. Go try

4:41

it out and support the show.

4:43

It's getgrist.com/self-hosted. Stop worrying

4:45

about the technology and use

4:48

the best. getgrist.com/self-hosted. Another

4:52

episode of self-hosted and here we are

4:54

talking about image yet again. The day

4:56

after we recorded the last episode, some

4:59

news about image drops, which we probably

5:02

should cover in this episode. Image

5:05

is joining forces with an

5:07

organization known as FUTO. Yeah,

5:10

and now the core team is gonna go

5:12

full time as a result of this. If

5:14

FUTO may ring a bell because they have

5:17

funded other open source projects, although this seems

5:19

like quite the commitment. It's a really

5:21

interesting situation. We find ourselves in

5:23

here. Essentially the gentleman

5:25

behind FUTO, I believe he made

5:28

some money with WhatsApp at some

5:30

point in the past. And

5:33

as such, he kind of got a little

5:35

bit burnt out with Silicon Valley and some

5:37

of the VC skull doggery that goes on.

5:40

And essentially his mission in

5:42

life now seems to be to

5:45

stick a middle finger up at big tech. At

5:47

least that's how I read it. Yeah, or

5:49

at least try to make an environment where

5:52

certain types of development that

5:54

might otherwise not be sustainable can be sustainable.

5:56

That's the more PC way to put it,

5:58

yes. Well, it does mean. There

6:00

are some transactions here, right? So

6:02

Fudo does now technically own the

6:04

trademarks and the source code. Although

6:07

the project will remain a GPL. So

6:09

I suppose if things went sideways, the source code is

6:11

still available. Yeah. The core team stays in place. This

6:14

is just going to be their full time job right

6:16

now, which is huge. I mean, you can't really understate

6:19

that. And the core team seems to be

6:21

providing the direction and the

6:23

feature set. With the only real

6:25

request that I can see that Fudo has made is that, you

6:28

know, build this whole thing so that way we can make a

6:30

cloud service out of it one day, which I think the image

6:32

team was already going to do. Right. As

6:35

I understood it, I've been talking to Alex

6:37

Tran. We've had him on the show before, but I've

6:39

been talking to him a lot in the last couple

6:41

of weeks about this news, you know, trying to make

6:43

sure that everybody's happy with

6:45

it. The reaction to this has

6:47

been pretty interesting because I

6:50

will certainly admit my heart dropped when I

6:52

read that first sentence, initially

6:54

thinking, oh God, image has been bought

6:56

out. But as you just said,

6:58

Chris, the source code remains a GPL. And I

7:01

think that alone, that single fact

7:03

alone gives me huge peace of mind

7:05

that even if we need

7:07

to rename image because of the trademark situation

7:09

in a few years, for whatever reason, even

7:13

if that needs to happen, the

7:15

source code and all the effort that's being put in here,

7:17

it's not going to go to waste. It does seem

7:19

like one of those situations where the benefits are pretty

7:21

clear. You know, the team that's been setting the direction is

7:23

going to be the same team that continues

7:26

to set the direction. They're not going

7:28

to change anything dramatically about the project. It's going

7:30

to remain a GPL. Those things, those

7:33

are obviously palatable benefits, let

7:35

alone getting the dev team hired. That's just, that's

7:37

huge. Yeah. And the downsides are all kind of

7:39

what ifs, like, well, what if in five years,

7:42

like they don't see eye to eye, like those

7:44

are all the downsides are like what ifs and

7:46

those are things that we could always pivot, tuck

7:48

and roll. I mean, it's a legitimate concern. This,

7:51

this gentleman who's behind FUTO, for example, what

7:53

if he gets bored? And decides

7:55

to do a Russ Haneman and drive off with

7:57

a Trethkomas into the distance. You know, it's like.

8:00

what's going to happen to the project in that situation.

8:02

But I think also it speaks

8:05

to the fact of how little trust

8:07

we have in tech

8:10

in general that most

8:12

people's first reaction to this is negative.

8:14

Yeah, I think, you know, people fool

8:16

me once. How do you think people

8:18

will react when they do put a price tag on it?

8:20

I think that's the goal here is Fudo is going to

8:22

say, Hey, if

8:24

you use this, why not pay for it? Like it's still free.

8:27

You can still get it. You can still self host it. But

8:30

if you'd like to send us 10 bucks, you can

8:32

send us 10 bucks. How do you

8:34

think that's going to go over? I think if you listen

8:36

to their live stream, they cover this speculation

8:38

outright and they just say, Yeah,

8:41

why is it okay for open

8:43

source software to be expected

8:46

to be free if it's going to be

8:48

of high quality and a

8:51

commercial alternative, a genuine

8:53

commercial alternative to something like Google Photos

8:55

or iCloud photos or whatever you use,

8:57

why should the expectation that

8:59

the fact that the software is free

9:01

as in open versus

9:04

free as in cost, that expectation

9:07

isn't based in the

9:09

capitalistic societies that we actually live in. I

9:11

mean, the very fact that we have open

9:13

source software at all is a minor miracle.

9:16

Why is it not okay to have these people be

9:18

paid to eat? That's how I take it anyway. I

9:20

hope they do. And I hope it's pay what you

9:22

want. Because if you asked me, I think image is

9:24

worth 100 bucks. I think every major

9:27

version could be worth 100 bucks. That's what we would have

9:29

paid back in the day for software like this. I actually

9:31

asked Alex this question directly. I said, what

9:33

is the monetization strategy? And he said, yeah, we're

9:35

probably going to charge a license fee or some

9:37

kind of a fee, a small

9:40

amount of money he said, I don't know what that actually means. It

9:42

could be 100 bucks. It could be $10.

9:44

Who knows? But I think they're

9:46

going to look at charging some kind of a fee

9:48

to use image as a one time thing. I don't

9:51

know whether it's going to be a semantic version, or

9:54

like a major version fee, like every time they

9:56

rev the major version, you pay again. To

9:58

me, that would seem to make the most in terms

10:00

of making the project sustainable. And that's

10:03

certainly what, if you look

10:05

at commercial platforms like iOS apps love to

10:07

do, and Blue Iris does this as well,

10:09

like every year you pay for updates and

10:11

stuff, I'd certainly pay 30, 40, 50

10:13

dollars without even thinking

10:15

for something like image. If

10:18

it's as good a quality as it is now, in

10:20

a couple of years or better, in a couple of

10:22

years time, sure. Yeah,

10:24

I didn't get the impression of a licensing scheme. I got

10:26

more of an impression that you could go, just do a

10:28

Docker poll and get it all up and running, but if

10:30

you wanted to go to the website, there'd be a pay

10:32

button. I think that really is

10:34

the right, it's essentially value for value is what

10:36

it is. You're just not calling it that. It's

10:38

let people show up and contribute the value they

10:40

think it's worth. And yes, good

10:43

software, good self-hosted software is worth paying

10:45

for. And I think the

10:47

model where it's all free, it's all available up like

10:49

on Docker Hub or GitHub or wherever they wanna host

10:51

it, and you can go deploy it and play with

10:53

it, learn to love it, then

10:55

it's worth a lot more to you all of a sudden than it was. If

10:58

you had to pay before you could even try it. All

11:01

right, I'll give it five bucks. Okay, I've heard good

11:03

things. I heard Alex and Chris talk about

11:05

this. I'll throw five bucks at it. But now

11:07

I've used it for more than a year. I'm like, yeah, I think six

11:09

things like worth a hundred bucks to me. Like, it's...

11:12

Yeah, exactly. I think

11:14

it could be a good model. Short-term charging for

11:16

the software, a bit like Ubuntu tried to with

11:18

their download slider, I think is the

11:20

goal. And then long-term as some kind of a hosted service

11:23

would seem to be the way to go, yeah. Would you

11:25

consider that? Like if it

11:27

was secure, if you were comfortable that

11:30

it was private, would

11:32

you consider using an

11:34

image hosted service because then you're getting

11:36

things offsite? I'm probably not

11:38

the target market for that service,

11:40

but I could see people like

11:42

me buying it for family members,

11:44

absolutely. Yeah,

11:47

I'm always a little tight on space. I've got a

11:49

maximum of eight terabyte I'm working with on

11:51

my home system. And I

11:54

could always try to add more solid state or

11:57

I try to reduce how much I...

12:00

I don't keep my full media library on there,

12:02

I keep an abbreviated version of it, but

12:04

I do keep my whole photo library on there. And

12:07

so that's eating up the bulk of the

12:09

space. And I could see myself

12:13

just saying, I'll just put that all up

12:15

on the image cloud, if say I trusted

12:17

the storage, going back to something

12:19

that, some other backup method that

12:21

doesn't eat that storage space on my O-Droid. I

12:23

could see myself maybe being a customer just because

12:25

storage space is tight, or definitely

12:27

for family members. If

12:30

the storage plans were reasonable and secure, I'd

12:33

probably pay a fee for that. Would you pay as

12:35

much as Google or iCloud charge? That's

12:37

trickier because in theory, you're getting a lot more for

12:39

that cloud membership, right? There's a lot of other

12:42

services. I mean, iCloud for two

12:44

terabytes is 10 bucks a

12:46

month. Yeah, I just don't

12:48

think two terabytes will quite cut it for me. And

12:51

it seems tricky because they can't really

12:53

do a lot of mass deduplication. So

12:55

they're really, it's probably a high storage

12:57

cost. Particularly if they're encrypting it in

13:00

any kind of meaningful way. They

13:02

can't do any really. I also

13:04

understand that this is gonna allow the

13:06

image team to accelerate plans for a

13:08

stable release on their roadmap. Oh,

13:13

that's great news. Yeah, I'm

13:15

all for that. I think too that getting

13:18

the software kind of nailed down, you would think

13:20

like if you wanna make a cloud service out

13:22

of this, even the next couple of years, gonna

13:25

wanna get to a stable version and then begin optimizing

13:27

the backend for that. Because you're gonna have to run

13:29

this thing at scale, so you better start fixing some

13:31

of these problems. Right, yep. So I

13:34

could see sort of a business driver need for that, but

13:36

also just the user base would probably really appreciate. I

13:38

think the other thing that FUTO is gonna give them is

13:40

the runway to hire people who have done that kind of

13:43

mass service build out before,

13:45

access to the correct developers

13:48

that have the right cloud experience and caching

13:51

and scaling things. It's

13:53

not a simple ask to just take

13:56

what should be a system that runs

13:58

on your local server. put it in the cloud

14:00

and open it up to tens of thousands of people, there's

14:02

going to be some pain as they do that. And I

14:04

think one of the things that

14:06

I'm sort of hearing from Alex is that

14:09

that's going to be possible now with this

14:11

collaboration with FUTO. I think calling it

14:13

a collaboration is fairly reasonable too. I

14:15

mean, so far we've

14:18

been given no reason other

14:20

than just general tinfoil hatness

14:23

to think that FUTO is anything other

14:25

than altruistic. They have a pretty

14:27

long livestream that you can find on

14:29

YouTube, and they go into some

14:32

of their motivations there, and I watch that, and

14:34

I buy it. I mean, I think the main guy

14:37

behind this seems like he's got

14:39

big ambitions and has

14:41

seen a few things. He also worked at Yahoo

14:43

in the Yahoo gaming area, and

14:46

he has seen a few cycles, and he's made all

14:48

the money he needs to make, and he's got enough

14:50

that he can set aside to kind of fund this

14:52

thing for multiple years before he even begins to pay

14:54

for itself. I mean, I like to think if I

14:57

was a billionaire, that's something I'd

14:59

do. I don't know though. Desert

15:01

Island sounds pretty appealing. Yeah, well, after you

15:03

get your desert island and your jet, and

15:05

your condo in every major city that you

15:07

enjoy, then you set aside a few bill

15:10

to fund your favorite open source project. You

15:12

might, you just might. Yeah,

15:14

absolutely. Now, another thing that popped up

15:16

in this whole news cycle was GrayJ,

15:18

which is an app from Lewis Rossman,

15:21

who is involved with FUTO, opening some

15:23

kind of a maker

15:25

space, like a repair space for the community

15:28

in Austin. And Lewis, as you

15:30

know, is big into right to repair and all

15:32

that kind of stuff. And GrayJ

15:35

is basically a way of

15:37

following individual creators that

15:40

transcends a specific platform.

15:43

And the example I think that Lewis gave when

15:45

he launched the app was Eli

15:47

the computer guy on YouTube still

15:49

uploads videos to other platforms, but

15:52

because he was de-platformed by YouTube,

15:55

lost a bunch of people and people don't know

15:57

where to find him anymore. And I think GrayJ

15:59

tries to. Now

16:01

one of the problems that they ran into was that

16:04

people accused them of not understanding

16:06

what open source meant. And

16:08

the reason why was because Greyj

16:11

was released with a license

16:13

that prohibited commercial redistribution as

16:15

they understand it. Because

16:17

they wanted to try and make sure that there

16:19

weren't a bunch of grey market clones of Greyj

16:22

appearing in Google Play

16:24

and app stores and stuff like that. To

16:27

circumvent some of the stuff they were

16:29

trying to protect against. And I don't

16:31

know, I feel like the people who are saying that

16:33

FUTO simply doesn't understand open source licensing are

16:36

a little wide of the mark. But

16:38

it's just something to be aware of that they've

16:40

got some skin in this game already. Yeah, I

16:43

guess I should have been paying attention a little

16:45

bit before now, but I'm definitely paying attention at

16:47

this point. I'm wondering

16:49

now what other projects they're going to go

16:51

after. I mean, you look at they've got

16:53

a video project with Greyj.

16:55

They've got another photo sharing feed

16:58

thing called Circles, which maybe

17:02

it's a little bit like

17:04

Facebook slash Instagram a little bit.

17:07

Like it's photo related, but they

17:09

say it's built with the security

17:11

model of Signal in mind. And

17:14

there's a bunch of other stuff to do with like

17:16

voice inputs and other privacy respecting

17:18

things. So it's

17:20

going to be fascinating to see what FUTO does in

17:22

the next one, two, three years. About

17:26

eight months ago, I set out to

17:28

answer the question, what is the perfect

17:30

media service CPU if such a thing

17:32

could possibly exist? And

17:35

I wrote a benchmark script in conjunction with

17:37

my good friend Morgan to find out. Because

17:39

as you probably know from listening to this

17:41

show, I'm a bit of a fanboy when

17:43

it comes to Intel's Quick Sync technology. This

17:46

is a piece of hardware circuitry that's

17:48

built into pretty much every Intel

17:50

graphics card, iGPU that's built

17:52

into your CPU since

17:55

2014, 15 sort of era. So

17:58

it's been around for a long. The time. The.

18:01

Was particularly interesting in a media

18:03

server context is a this hardware

18:05

transcoding uses five to ten watts

18:08

to trans code multiple for case

18:10

streams faster than you do it

18:12

in software and so what I

18:14

wanted to do with figure out

18:17

what is the difference between. That.

18:19

Say a seventh or eighth Jen

18:22

Intel processor, and some of the

18:24

more modern twelve thirteenth, fourteenth Jen

18:26

processes compared to the super low

18:29

power in one hundred type stuff.

18:31

And so that's what we've done.

18:34

Okay, kind of looking for a baby.

18:36

The best performance to price to power

18:38

usage. Sweet spot. Yeah proper. it depends

18:40

on the coming up so be warned

18:42

to assess Afghanistan read: didn't like it

18:45

at the weekend There was one particular

18:47

guy that say I decided to replace

18:49

my reddit thread with conclusion. Well it

18:51

depends on he goes well. What was

18:53

the point in the entire? Test them

18:56

on my feet. read the article days

18:58

they're probably not. Say. The

19:00

conclusion which I'm straight to that is

19:02

that you might find this month. And

19:05

eleventh Jen into Cp You and motherboard com

19:07

by was the best one you can find

19:09

on E Bay. Next. Month's

19:11

it might be eight Jan and a month after

19:13

it might. He might decide well as the I

19:15

want Ddr five or something like that, In which

19:17

case you can only go thirteenth gin or newer.

19:19

For. You might say to

19:21

yourself, water will not great the motherboard

19:24

and city. What about Arc teepees so

19:26

as get into some of the results.

19:28

Now one of my wonderful contributors to

19:31

this quick sink. talc repo sort

19:33

of mention morgan wrote probably write the

19:35

script and there's another user code alice

19:37

sumo i hope i got that right

19:39

to said a wonderful scripts which helped

19:41

visualize some of this data so some

19:43

a so you check out the so

19:45

nice because they'll be a link to

19:48

the blog posts were all of this

19:50

stuff gets on taxes grass in mass

19:52

chart so that kind of stuff's but

19:54

hero a visualization to in python which

19:56

helped kind of give us a pretty

19:58

good idea that the spread across generations.

20:00

So what's interesting about quicksync,

20:02

if you look at some of the

20:05

older chips, so let's

20:07

take a Skylake chip, which is

20:09

Intel's 6th gen. That

20:12

has H264 hardware

20:14

decoding in it, but 5th

20:16

and 4th and 3rd and so on, they

20:19

don't really have very much in the way of

20:21

hardware encoding whatsoever. I mean, they've got some basic

20:23

MPEG-2 hardware stuff, AVC. There's

20:25

a graph, a fantastic graphic on

20:28

Wikipedia actually, which shows you how

20:30

over the years they've added new

20:32

codecs and stuff like

20:34

the 13th gen has AV1 support, for

20:36

example. So if you want the most

20:38

modern codec support, buy a new chip

20:40

because the downside of it being

20:42

a hardware encoder is it's a

20:44

very specialized piece of circuitry, like an

20:47

ASIC, that can only do one thing

20:49

and it won't do

20:51

anything else. You can't add AV1 support in

20:53

hardware to a chip from five years ago.

20:55

Yeah, it's what it ships with is what

20:57

it's always going to support. Exactly. Now

20:59

quicksync is great, as I mentioned, because of

21:02

its power efficiency. So, you know, you're doing

21:04

multiple 4K streams at sort of 5 to

21:06

10 watts of power consumption to

21:08

encode a 4K file. Sometimes

21:11

at more than 1x speed with some

21:13

of the 10-bit HEVC files, the H265

21:15

stuff, which is

21:18

almost unheard of, to be honest. But

21:21

let's jump into some of the results specifically.

21:23

So if we want to look at, say,

21:26

there's a chart in the blog post linked, as I

21:28

mentioned, if we want to compare, say, an Intel i58500,

21:31

which just happens to be the CPU that's

21:33

been in my primary media server for the

21:35

last three, four years, something like

21:37

that, to an Intel 13th

21:40

gen, so an i5-13600K, which also just

21:42

so happens to be the CPU I've

21:44

just thrown into my primary

21:46

media server downstairs in the last

21:48

couple of weeks, You would think

21:50

there would be an absolutely enormous difference between

21:52

these chips. An I58500 was released in 2018.

21:54

A 13th Gen was released. Last

22:00

year, so was that for years. For.

22:02

We're for your five years. six

22:04

years even have Cp progress. You

22:06

think that he would seem to

22:08

the transcoding for me, my care.

22:10

a tenancy p file with ties

22:12

to Six Vortex if you file

22:14

would be orders of magnitude different.

22:17

When you surf, when it comes

22:19

to Cp, you encourage. That is

22:21

the case. So with a a

22:23

thirteenth Jen, you're looking at about

22:25

thirty seconds to code this specific

22:27

father. I include a spot benchmarks

22:29

and with the eighth Jen, it's

22:31

about sixty seconds. So. It's okay, the

22:33

grunts of the software encode says they're

22:35

harboring toting involved the new cities are

22:37

lot faster and the he's about the

22:39

same amount of energy to encode that

22:41

five as well as they do it

22:43

so as you doing software encodes but

22:45

it's a no brainer but if we

22:47

jumped straight into the hardware eight to

22:50

sixty four, ten a t P and

22:52

code. This. Thing. It

22:54

may be has a two second difference. In

22:56

the haven't touched. The.

22:59

Base to six for encoders What?

23:01

So ever since. Twenty Eight

23:03

seed is what we can extrapolate from that.

23:06

I gotta work and Good Math says exactly

23:08

as As and things are a little bit

23:10

different when he gets a forte. The gap

23:12

goes from being one or two seconds to

23:14

being five or six seconds, but it's still

23:17

in Nxt six fall and it doesn't matter

23:19

what Cpv get, as long as it's seventh

23:21

or eighth gen newer. The performance is the

23:23

same, even despite the fact that they have

23:26

completely different graphics chips or this. The A

23:28

Stand has H D Six Thirty graphics. where's

23:30

the thirteenth has you? H D Seven seventy

23:32

with different clock. Speeds another estimate, but

23:34

as we discussed. The. Hardware circuitry

23:37

is set in stone when it leaves

23:39

a factory. And it looks

23:41

like the to six for circuit has

23:43

been set in stone since Twenty Eight

23:45

Team. Now let's jump into H To

23:47

Six Five territory a T V Cu.

23:49

It's think again that they would be

23:51

a huge gulf between. The.

23:54

Eighth and Thirteen ten ships. But it's it's

23:56

just not the case. So the eight ten

23:58

Cp you takes about. Forty five seconds

24:01

to encode the x two, six

24:03

by file and eight bit whereas

24:05

the thirteenth Jen took. Forty.

24:08

Seconds Thirty nine seconds I think is was

24:10

the average result in the in the tests.

24:12

Things. Do get a little different though

24:14

when you jump up to the ten

24:16

bit Forte file. This is the file

24:19

that really sorts the men from the

24:21

boys. This is the one that makes

24:23

everything com to it's knees. Now the

24:25

apes. Jen Cp You take nearly a

24:27

hundred and eighty seconds to encode this

24:29

file whereas the thirteenth Jen takes a

24:32

hundred and thirty seconds or so you

24:34

just a little over two minutes. This

24:36

is nearly three minutes with the other

24:38

chip, so there is a big difference

24:40

with specifically for K. Ten

24:42

bit, eight to six, five files.

24:44

but every other test the I

24:46

ran. Didn't. Really matter. Again,

24:50

so if you're doing really the latest like we're

24:52

talking age to sixty five for gay ten

24:54

bit video. And. It, It's going to make

24:56

a difference if you're doing a lot of that. But. Outside

24:58

of that, these encounters are basically. Pretty.

25:01

Much set in stone and kind of to your

25:03

earlier. Conclusion. It would make

25:05

sense that. The. Most latest Kodak with the

25:07

most latest video technology that's gonna have the

25:09

best support in the most. Latest.

25:11

See Buba once they in theory nail

25:13

a to sixty five four k. Probably.

25:16

Not going to change much for quite

25:18

awhile now. Spinner: the these platforms are

25:21

a quite different The Thirteenth Jen supports

25:23

Pc I huge and Five. It has

25:25

Ddr Five memory which is a lot

25:27

faster than Ddr for the eighth Jen

25:29

chips and I'm I'm deliberately comparing these

25:31

eight vs. the Thirteenth because everything in

25:33

between is sort of a sliding scale

25:35

does to give you will an idea

25:37

he would think that the speed of

25:39

a memory in the system would make

25:41

it does more. I'll bet more Did

25:43

a more of a difference like made

25:45

as more. you know ltl for each

25:48

other no cash on this on the ship

25:50

or something that makes a difference now doesn't

25:52

seem to matter it's like these little encoder

25:54

accelerator chips are just little factories and the

25:56

factory staff remain the same and the capacity

25:58

remains the same and The world goes

26:00

on around them. Now, of course, most

26:03

of us are running these sorts of chips

26:05

in an environment where we're doing things like

26:08

a bunch of containers, we're doing a bunch

26:10

of VMs maybe on these systems. That's

26:12

where you'll start to notice more PEP in

26:15

the system with a newer chip and stuff

26:17

like that, again, down to the things like

26:19

more modern memory bandwidth and more M.2 lanes

26:21

and all that kind of stuff. But if

26:24

all you're doing is a bog-standard, super simple

26:26

media server that's just serving a few files

26:28

for Jellythin with the odd transcode here or

26:30

there, I don't see any reason to go

26:33

for anything other than sort of 7th, 8th, 8th gen

26:35

to be honest with you. Now,

26:37

we do need to talk a little bit about

26:39

Intel ARC whilst we're here because I was lucky

26:41

enough one of the podcast listeners, I

26:43

am Spartacus, he goes by, sold

26:46

me an ARC Pro A40 GPU. It's

26:48

an OEM-only card. You can't buy it

26:50

on the market, so to speak. It's

26:52

sort of a system integrators-only

26:55

card. This thing is an

26:57

absolute monster. So you remember how

26:59

I said the 8th gen chip took nearly 180 seconds to encode

27:02

that 10-bit 4K file? Yeah. This thing

27:04

took about 38 seconds. Oh,

27:08

that's encouraging. Now,

27:11

we've got to temper your excitement a little

27:13

bit because it used 65 watts

27:15

whilst it did it, not 5 to 10

27:17

watts. Oh, right, of course. Right,

27:19

of course. But... Well, it just depends on what you're optimizing

27:21

for, I suppose. Well, yeah, exactly. It's like when you're running

27:23

in the rain. Do you get wetter when you run in

27:26

the rain or walk in the rain? Like, if

27:28

I'm running a task for 38 seconds instead

27:31

of 3 minutes, am I cumulatively

27:33

using more energy even though it's,

27:35

you know, like 6 times the amount of

27:38

energy being used or less? Like, is it

27:40

better to have that heat dumped into the

27:42

box in one short burst or

27:44

have it spread out over 3 minutes? Really, only

27:46

you can answer that question. I just

27:48

thought it was really interesting to have an ARC card to be

27:50

able to test. It was also pretty

27:52

interesting as well because the HL15 that 45Drive sent me a

27:55

few months ago had

27:59

what I can only describe... of potato CPU and

28:01

it had the Xeon bronze chip, the model number

28:03

escapes me, but it was like a 2.2

28:06

gigahertz 2018 sort

28:08

of chip, it was not fast, it didn't have very

28:10

many cores. So with that Xeon

28:13

bronze 3204 CPU which ran at

28:15

1.9 gigahertz, I must correct myself,

28:18

the ARC Pro ran at, it

28:20

took 140 seconds to

28:22

do that 4k 10-bit in code.

28:25

When I threw the ARC Pro into my epic

28:27

7402 build that I did recently,

28:29

that dropped to 39 seconds.

28:32

So 101 seconds quicker just

28:34

because of the CPU. Yeah, wow.

28:38

So it does make a big difference. Yeah and I

28:40

could see if you're in a situation where you get

28:42

your hands on an ARC GPU and

28:44

you don't mind the power use, it's still

28:46

cheaper, cheaper as far as I

28:49

guess energy usage than probably

28:51

like you know systems that we were building five, ten

28:53

years ago that were probably pulling you know

28:55

five, six hundred watts from the wall continuously. Easy.

28:58

So in the grand scheme of things, 60 watts

29:01

isn't horrible compared to what we used to do,

29:03

especially if your workload is enough

29:05

where something like that would actually be beneficial.

29:07

What I would love to do is pull

29:09

in some experts to get NVENC benchmarked in

29:12

there as well as some of the AMD

29:14

APUs that exist as well. So if

29:16

that floats your boat again there'll be a link

29:18

in the show notes to you know the repo

29:20

where you can contribute to the little bit janky

29:22

bash script but we'll get a life through together.

29:27

unraid.net/self-hosted. Go check it

29:29

out. Unraid is powerful

29:31

and flexible from a local file

29:33

storage system, maybe a gaming rig,

29:35

content creation and everything in between.

29:38

Unraid allows you to get the most out of

29:40

your home lab and self-host all your important data

29:43

and your important services. It's

29:45

a network attached operating system that lets

29:47

you mix and match. You got a

29:49

closet full of old drives? Say no

29:51

more my friend. You can take those

29:54

mix and match those drives and securely

29:56

store and share your data. Manage your

29:58

VMs and your Docker applications. all

30:00

from one nice web-based interface. One

30:03

of the features we love is Unrade lets you easily

30:05

pass through a graphics card to enable gaming

30:08

or accelerated VM or container, maybe even want

30:10

to video edit on your Unrade. It is

30:13

that flexible. And Unrade's UI

30:15

is so simple that managing an

30:17

entire stack of Docker containers is

30:19

simple, straightforward and easy to

30:21

figure out. And they have new applications from

30:24

hundreds of user-created templates in the community app

30:26

section all the time, something you should always

30:28

be checking out. It makes it really easy

30:30

to tuck and roll depending on your expertise

30:32

level from total beginner to expert

30:34

who wants the most from their system.

30:36

It's a powerful user-friendly operating system for

30:39

home servers. Go build the perfect storage

30:41

solution with the hardware you already have.

30:43

Use Unrade OS. Just support

30:46

the show by going to

30:48

unrade.net/self-hosted. That's unrade.net/self-hosted.

30:51

And go check out Unrade, a powerful, easy

30:54

to use operating system for servers, for storage

30:56

and for your applications. Use

30:58

your hardware with unmatched flexibility and go

31:00

say goodbye to that closet full of

31:02

mismatched drives. We've all got it. No

31:05

shame, but Unrade is the solution.

31:08

Check it out and support

31:10

the show. Go to unrade.net/self-hosted.

31:14

So if we're okay accepting that anything newer

31:16

than a sort of seventh or eighth gen

31:19

CPU is the future for

31:22

media servers specifically, we've

31:24

got to look at the rest of the ecosystem

31:26

around the chips because it's all very well and

31:29

good saying, yes, I can buy an eighth gen

31:31

i5-8500 for 60 bucks on eBay, but

31:33

it's no good if you can't find a motherboard. And

31:37

these chips are not spring chickens.

31:39

And by extension, that means the motherboards have gone

31:42

the same way too. So it

31:44

can be quite tricky to piece together

31:46

an entire system. So you've

31:48

got to find, I mean, DDR4 RAM, I

31:50

mean, that's been around for a decade and

31:53

its performance characteristics are pretty well understood

31:55

at this point. So that's not really

31:57

a concern. DDR4 memory is a commodity

31:59

at this point. and it's cheap. DDR5

32:01

is about twice the price still,

32:04

which is unfortunate. But let's

32:06

say you want to find an 8th gen motherboard, you

32:09

can find a lot of used gaming

32:11

grade motherboards, and for some people that's

32:13

totally fine. Pair that with a Pi

32:15

KVM and you get all pretty much

32:17

the same features as a built-in

32:19

BMC chip would give you on a more

32:23

server grade motherboard, for want of a better

32:25

word. But the trouble is, those server grade

32:27

boards, the BMC integration for the

32:29

most part is just that little bit

32:31

more reliable. And for me, as

32:34

you well know, I've got my backup server in the UK on

32:36

the other side of the ocean. I can't

32:38

deal with anything other than perfect reliability

32:40

on that box really. And

32:42

the IPMI has bailed me out a couple of

32:44

times after like power outages in the remote country

32:47

and stuff like that. So as much

32:49

as I love advocating for Pi KVMs when you

32:51

can go down to your basement

32:53

and physically touch the box and finagle it back

32:55

to life again, when it's a

32:57

remote deployment or you don't have easy access to

32:59

be hands-on, I'll still advocate

33:01

for the BMC stuff. And that's where things start to

33:04

get a little bit more tricky. So to give you

33:06

an example, the LGA 1151

33:08

socket, which is the one that the

33:10

8th gen Intel CPUs use, if

33:13

you want to find a motherboard with BMC, the cheapest

33:15

I could find was sort of in that $200 to

33:17

$300 range on eBay, which, you

33:20

know, some of the advantages of a cheap CPU

33:22

start to evaporate at that point. Yeah.

33:25

Although, you know, if you factor in the cost

33:27

of Pi KVM, which probably about a hundred and

33:29

something bucks, it's not so bad. You know, you

33:32

start to think about if you were going to

33:34

put a Pi KVM on there, it's $120. That's

33:36

true. Yeah, I hadn't considered that. Because, I mean,

33:38

in my mind, the Pi KVM has sunk costs,

33:40

but that's not fair to assume that at all,

33:42

is it? Because you might be buying a new

33:45

one for this. Right. Yeah.

33:47

Good point. I am a big fan of

33:49

Pi KVM. I have used it, you know,

33:51

remotely in anger and been, you know, perfectly

33:53

happy with it. But I also agree that if

33:55

you can have built in management tools as well in

33:58

a situation where it really be a

34:00

lot better off. Although I wonder,

34:02

what about, okay Alex, just going down this route,

34:05

what about having like two machines? And one of

34:07

them is just like a little one liter PC

34:09

that's a remote box that you could do support

34:11

from. Great question. Well I covered a

34:13

little bit. There were some test results that came in with

34:16

the N100 chip, which is basically

34:19

the efficiency cause of an Intel

34:21

12th gen only. So

34:24

for those of you that aren't familiar,

34:26

Intel in the, I think it was

34:28

12th gen, introduced their new P and

34:30

E core architecture. And they released this

34:32

really interesting chip called the N100. There's

34:34

also a slightly bigger brother called the

34:36

N305, which is just a slightly beefier

34:39

version as being like the ultimate conclusion

34:41

of their Atom series. So

34:43

this thing, the N100 sips, absolutely

34:45

sips power. It uses about

34:48

six watts idle for the entire

34:50

system, which is quite something.

34:52

The downside is of course, it's a bit

34:54

of a potato when it comes to doing

34:56

anything processor intensive. So you

34:58

start thinking to yourself, well why don't

35:00

I look at these small form factor

35:02

PCs, ignoring the fact that N100 motherboards

35:04

are difficult to get hold of, often

35:06

don't have enough SATA ports. And

35:09

sometimes you end up having to do weird stuff to get

35:11

things like 10 gig on them and stuff like that. Because

35:13

they just don't have enough PCIe lanes

35:15

to be useful. I think they have about nine

35:17

lanes to go around the entire system, which just

35:19

isn't enough. So you think to yourself, why don't

35:21

I use a small form factor PC like some

35:23

of those Dell ones that you talk about all

35:25

the time, Alex? And it's a great point.

35:28

If you have a big honking storage

35:30

box that you're happy with, you'll

35:33

probably be in the red

35:35

for quite a long time, energy efficiency

35:37

wise, if you want to buy a

35:39

new system, it's hard to

35:41

justify the ROI on buying a new

35:44

system versus you

35:46

look at some of the sales that Dell has

35:48

on these boxes. An i5 6600T box, I

35:52

picked up for $138. Wow. With 8 gigs of RAM and 128 gig SSD, for

35:54

example. That's not bad.

36:00

And they have QuickSync. Yeah, little

36:02

QuickSync box. These T processors,

36:05

I never really liked them. A lot of

36:07

people go, oh, but they're low power. They're

36:10

not low power, they're just capped at a

36:12

certain TDP. So at the

36:14

lower end, you've got all the same grunt, for

36:16

the most part, as a desktop

36:18

class chip without the T designation.

36:21

But all T does is it just cuts it

36:24

off with the kneecaps. Haha, you can't go any

36:26

further than this. The T stands for capped TDP,

36:28

is that what it stands for? T

36:31

stands for totally annoying because

36:34

they're often more expensive to buy because

36:36

they're more desirable because people think they're

36:38

a low power chip. Even

36:40

though idle, they're exactly the same. These

36:43

small form factor builds, they make sense because there's only

36:45

so much heat you can get out of a little

36:47

one liter box, for example. But

36:50

if people are buying them to put

36:52

in a full size chassis with a

36:54

proper cooler situation and stuff, I've never

36:56

really understood the benefit. Yeah,

36:58

I guess that makes sense. I could see

37:00

that as a line. I do think too, at

37:02

that price though, you could almost afford to

37:04

have a backup one in a drawer

37:06

and then you could throw these things in just the

37:09

worst conditions. And if

37:11

they burn themselves out, you just pull it out and

37:13

pop in the other replacement that's in the drawer. It's

37:15

kind of the Raspberry Pi strategy. I was going to

37:17

say, you said the same thing about a Raspberry Pi

37:19

five years ago. Yeah, I think that's where

37:21

we're at now. It's

37:23

assuming though that you have storage somewhere else, I suppose.

37:26

Yeah, I mean, you can get an eight terabyte, two

37:28

and a half inch SSD if you want. I

37:31

mean, that's a lot of Rick and Morty right there. That's

37:33

what I do in my little home Odride setup, right?

37:35

Is I'm just, I'm hanging an eight terabyte SSD off

37:38

the thing. Or I think

37:40

it was Jeff Gailing this week just highlighted

37:42

that you can get these carrier boards now

37:44

for the Raspberry Pi five that

37:46

have four M.2 slots on them.

37:50

There's a bunch of other small low powered sort of NAS boxes

37:52

that are flash only. And

37:54

since the last time we had a discussion

37:56

like this, the landscape has

37:58

changed in... terms of flash

38:01

prices versus spinning rust like yes

38:03

twenty terabyte drives exist now but

38:06

so do four terabyte m dot two

38:08

drives at a reasonable price

38:10

if you can wait for a sale you know so where

38:14

does this leave us chris well

38:16

tell you what i like about the intel

38:18

stuff either the quick sync or using the

38:21

art card i guess unless

38:23

i'm wrong it seems like it's it's probably

38:25

a little simpler than any of the other

38:27

gpu vendors because it's all just really built

38:29

into linux now yeah maybe with the and

38:32

exceptions but i would imagine you're just

38:35

passing a device path through in a doctor

38:37

compose that points to like that d r i

38:39

device or something and then the applications

38:41

in the container just see that is that how it

38:43

works for like a practical standpoint to use the art

38:45

video card yeah same thing so the

38:47

quick sync stuff built into the c p

38:49

u shows up as slash there

38:52

slash d r i slash render d

38:54

one two eight and the

38:56

art gpu shows up as d one two

38:59

nine aha okay and then

39:01

so you could just map that to the device

39:03

in the container or it may be just path

39:05

them all through i suppose absolutely and

39:07

you could use could use both simultaneously

39:10

i haven't tried that i imagine it

39:12

would be fine but it would be up

39:14

to the ap i suppose to have some way of

39:16

scheduling those jobs on different places

39:18

yeah yeah that would probably be

39:21

the trick what would be ideal and

39:23

this this would go for a small form

39:25

factor like farm would

39:27

be if if something like jelly fin i know

39:30

this kind of exists for plex although

39:32

i never got it working was

39:34

some kind of like kubernetes based

39:37

transcoding remote worker

39:39

situation so you could

39:41

you could spin up pods for example and

39:44

have jelly fin point at those pods to

39:46

do processing a little bit like

39:48

we talked about in the last episode with image in

39:50

the machine learning stuff if you

39:52

could point at a remote endpoint and

39:54

say here is your compute go load

39:56

balance yourself maybe that would be a way to

39:58

do it That really should

40:00

be the way that is

40:02

the solution to getting on-premises,

40:06

LLMs that can work at scale and things like that.

40:08

If you could somehow have sort

40:11

of like take the Wyoming protocol that

40:13

Home Assistant has created for voice assistants

40:15

to be able to remotely communicate with

40:17

processing endpoints, if you could extend

40:19

that to all sort

40:21

of jobs of that like in-code jobs

40:23

or compute jobs that could be

40:26

approachable by folks like us that maybe you just run an

40:28

application in a docker container that points at

40:31

a hardware device and has a listening port

40:35

and then they use some sort

40:37

of auto-discovered DNS protocol. They find each other

40:39

and you can just submit jobs to it.

40:41

It would be nice. You could see something

40:43

that really is large at scale and

40:45

it's never going to happen but you know who could do this is

40:48

Apple. Apple could just build this

40:50

into all their devices with all of their

40:52

neural processors and you could just run a

40:54

local LLM that just runs across all your

40:56

iDevices at once. They'll never

40:58

do it but if anybody could they could. Indeed.

41:02

Yeah so there you go folks. That

41:04

is the perfect media service CPU is

41:08

up to you really. I know

41:10

that's such a cop-out to not pick one but

41:13

it really does depend on what's

41:15

on eBay this month. I think you'd be

41:17

too hard on yourself. I don't know. Because I think

41:19

it's actually really good information. If you start with the

41:22

eighth gen you're going to be okay and

41:24

anything later the better. It's true. Nothing like

41:26

this existed publicly for many years and I

41:28

actually pitched this article to Ask Technica sort

41:30

of three or four years whenever I did

41:32

that Google Photos thing for them and they

41:35

weren't interested. So it's been on my mind

41:37

for a very long time this and

41:39

a huge thanks to the community that offered

41:41

up their hardware to run these tests. I haven't counted

41:43

how many but it's got to be approaching a hundred

41:45

or so as a sample

41:48

size. Feel free to if you haven't

41:50

done it already. Just download

41:52

the git repo it runs a script it

41:55

pulls down some stuff from Linux S3 storage.

41:58

Takes about five minutes to run. run the tests

42:00

and then submit the results

42:02

to the GIST. I have

42:04

a GitHub action that automatically updates the

42:06

graphs based on those GIST results

42:09

every, I think I've set it to run

42:11

every week, so by all means,

42:13

I know I've drawn some conclusions today but the results,

42:15

you can just keep them coming. tailscale.com/self-hosted.

42:20

Go get 100 devices for

42:22

free. I

42:24

have Tailscale on every single system,

42:27

my mobile devices, my set-top boxes,

42:29

my family's computers, my containers even

42:31

have individual tail net nodes, and

42:34

I'm still not using all 100

42:37

devices. Tailscale is the easiest

42:39

way to connect your devices and your

42:41

services to each other directly wherever

42:44

they are. That's the idea. Secure remote access

42:46

to your systems when you need them, even

42:49

if they don't have the Tailscale client

42:51

on them, and it's fast, like really,

42:53

really fast. And Android

42:55

users, today is your day. Tailscale

42:59

is relaunching the app for Android, redesigned

43:01

from the ground up for a better

43:03

look and feel, faster performance, and even

43:05

more features. You can find out the

43:08

details at tailscale.com/blog. That's super

43:10

nice to see. I will also give a

43:12

mention for the Tailscale Apple TV app, which

43:15

is a game changer. It's absolutely mind-blowingly

43:17

awesome to have Tailscale on the Apple

43:19

TV, even as just like a node

43:21

on the network, but also

43:23

then to get access into a private network

43:25

of content and media that I have, and

43:27

when I move that Apple TV physically, it

43:29

still has the same tail net IP. It

43:32

can still access the same content. It's

43:34

an extremely powerful idea. I put everything on

43:36

my tail net now, so I have no

43:38

inbound ports on my firewall, and everything can

43:40

find each other on my tail net network,

43:42

including I even have a DNS server on

43:45

there, and I'm pinging and talking to everything

43:47

by name these days. It's

43:49

like my own genuine little internet that

43:51

I've built, and it's all protected by

43:54

wire guard, zero config, no fuss, machine

43:57

to machine talking directly to each other,

43:59

and it's perfect enterprises, you can cut

44:01

down that complex infrastructure. They have ACL

44:03

policies to securely control access to devices

44:05

and services with next gen access controls,

44:08

a dashboard that now supports dark mode,

44:10

and they have all the tooling you'd

44:12

need to integrate with your existing authentication

44:14

infrastructure. So for a home labber, you

44:16

know the 100 devices that

44:19

might just cover you. Go to tailscale.com/self

44:21

hosted. Then if you want to take it

44:23

to work one day, they have got solutions

44:25

that make this work and sing in the enterprise.

44:28

Simple secure networks for a team or

44:30

an individual of any size built on

44:32

top of WireGuard. Now with a

44:34

brand new rebuilt Android app. Try

44:37

today, support the show, and get

44:39

it for free on 100 devices

44:41

and three users at tailscale.com/ self

44:43

hosted. I

44:46

thought we weren't into dashboards. I thought

44:48

we decided that dashboards were crafty and

44:51

pointless and old hats, no? Definitely

44:54

not into dashboards, Alex. Definitely

44:56

not, but I have

44:58

been kind of glancing at this muffle

45:00

dashboard. I actually know how you say

45:03

it, M-A-F-L. And

45:05

it's supposed to be a really simple, you

45:08

know, a couple of lines of yammer. You got

45:10

yourself a dashboard where you can track

45:12

services, get links to things, keep it all local.

45:15

I'm putting this out there because here's what

45:17

I realized while traveling I would have liked

45:19

to have had. I would like

45:21

to just open up my browser and

45:24

have a homepage that has a

45:26

links to, I have services on

45:28

multiple lands, not both, just

45:30

on my tail net. And

45:32

I would like to kind of know what things are up or down. Just

45:35

red light, green light. That's all I want

45:37

is like a logo, red light, green

45:39

light. And if I click it, it takes me

45:42

to the thing. And maybe I can have sections

45:44

so I can have a by location, studio, you

45:46

know, jukes, cloud stuff, I don't know. And

45:49

I think I'm getting close with this.

45:52

I'm not looking for something that's gonna be like the thing

45:54

I go share on Reddit with screenshots. And I don't want

45:56

it to have like all this

45:58

crazy information. I just want something. something that

46:00

will load fast, like I

46:02

put on my tail net, and then I

46:04

kind of get a once overview of everything that is

46:06

like my back end stuff. What

46:09

about the entropy that we discussed last time we

46:11

brought up dashboards? How are you going to keep

46:13

it up to date? I don't know. I

46:17

feel like this is where it's always going to fall down.

46:20

And I don't know why these dashboards often

46:23

don't let you edit their config inside the dashboard,

46:25

which seems like a quick way. It's like going

46:27

there. I'll just go edit this

46:29

really quick here and save it. No, no,

46:32

you got to SSH in. You got to go

46:34

to the file. It's like, come on. It makes

46:36

it take forever. There was a version of another

46:38

dashboard. I think it was called Flame that let

46:40

you do a dashboard like

46:42

this in a similar way to you

46:44

did traffic. So you put labels in

46:46

your compose file, and it generated

46:49

the dashboard from there. It

46:51

kind of made my compose file a bit

46:53

messy because it needed three or four or

46:55

five labels per container or something. Maybe

46:58

that's a way to do it. I'd also like to hear

47:00

what the audience has to say. Yeah, me too. Do a standard email

47:02

in and let us know. Speaking

47:04

of baboose, A-Ron came in with

47:07

100,000 sats. You're

47:09

our baller this week. That's for sure. Thank you very,

47:12

very much. And they wrote, I

47:14

just got offered a job that's a big step

47:16

in my career, and I wanted to share the

47:19

love and say thank you. This new position will

47:21

be DevOps engineering, and a large part of that,

47:23

DevOps knowledge and interest, is because of this podcast.

47:25

I wouldn't have found my love of automation if I

47:28

didn't discover Jupiter Broadcasting. From the bottom of my heart,

47:30

thanks guys. Well, thank you very

47:32

much, A-Ron. What's funny about this boost

47:34

is that I'm pretty sure

47:37

I sent you, Chris, a very similar

47:39

email five, six, seven

47:41

years ago before I was involved with JB, saying

47:43

thank you for all the shows. I

47:46

really appreciate messages like this because it's paying

47:48

it forward, and it's just nice to hear

47:51

that we heard it

47:53

from several people at conferences as well. And

47:55

I'm feeling pretty recharged from the audience lately,

47:58

so big thanks, guys. It is

48:00

one of the nicest messages you can get. Thank

48:02

you. And also congratulations, A. A. Ron.

48:04

Yeah. That's really great. A shaft and spanner came

48:06

in with 20,000 sats. Redis has

48:09

changed. You like that? Okay.

48:11

Funny name. It is a good one. Shaft and spanner.

48:13

I like it too. Redis changed their license in March,

48:15

they write. Home Assistant is now owned by the Open

48:18

Home Foundation and Image has transferred its

48:20

ownership to FUTO. Is this worthy

48:22

of an episode to discuss ownership, licensing, and

48:24

the long-term survivability of FOSS? As

48:27

we discussed about FUTO earlier, they

48:29

are not touching the license of

48:31

image specifically. So even

48:34

if they turn out to be bad actors, as

48:36

we discussed, we don't necessarily have any reason to

48:38

believe that they will yet. The

48:40

code's going to be fine. The project's going to

48:42

be fine. I think Home Assistant is another good

48:45

example of it's all fine as far as I

48:47

can tell right now. Like

48:49

the proof will be in the pudding. The

48:51

only juror that we can trust is time

48:54

here. The

48:56

fact that both of these projects have made

48:58

these moves, Home Assistant in

49:00

particular, being the founders of this

49:02

Open Home Foundation, hats off.

49:06

It's big picture thinking. It's thinking,

49:08

how do I make these projects sustainable

49:11

in the long term? I'm

49:13

sure Powerless one day will get bored of Home Assistant,

49:15

hopefully not in the near future but one day, the

49:17

guy's going to want to retire or something, I'm sure.

49:21

How do these people make sure that these

49:23

projects transcend themselves in the

49:25

long term? For these projects that

49:27

are hopefully going to be around for the rest

49:30

of our lives, like Home Assistant, I think it's

49:32

really important. Yeah, plus one to all

49:34

of that. I completely agree. We are obviously keeping our

49:36

eye on it though, Shaft. So if anything does look

49:38

a little funky or smells a little weird, we'll

49:41

be on it. But we're watching.

49:43

The reddish situation and the terraform

49:45

situation, I think are different. Maybe

49:49

not really. They're both primarily commercial

49:51

tools that were built on open source and

49:53

then were rug pulled. So

49:56

I think they are perhaps worthy of a different discussion

49:58

but I wouldn't lump them in with Home Assistant. and

50:00

image personally. They

50:22

have some pretty nice pricing plans. I

50:24

was speaking with Alex, who is the lead

50:26

developer of image. I know this is on

50:28

their roadmap for some indeterminate point in the

50:30

future, some like hosted version of image, like

50:33

that's going to be their business model, I

50:35

believe moving forward. But

50:37

the ente stuff, like you can buy

50:39

storage like you would buy iCloud storage

50:41

and yeah, you don't

50:43

need to worry about hosting yourself. So as

50:45

long as you trust where this data is living,

50:48

and I haven't dug into where it lives, I'll

50:50

be honest, but as long as you trust where

50:52

it's living. Ozzy

50:55

comes in with 17,345 sats. Have

50:58

you guys thought of self hosting a Bitwarden instance

51:00

and then back ending it with tail scale so

51:02

it's available everywhere. Enterprise licenses

51:04

are a bit expensive, but you could

51:07

also do it with six free family accounts with each

51:09

license. Yes, I have and for much the same reason

51:11

as I don't run head scale. There

51:13

are just a couple of things in my infrastructure that

51:15

are so critical that I can't

51:17

afford for them to go down like if my

51:19

Bitwarden disappeared, I

51:22

would be I should probably back that

51:24

up somehow. Now I'm talking about it

51:26

in the episode. Yeah, Voltwarden, I

51:28

think they just rename themselves again, something

51:31

else. But yeah, Voltwarden is the self

51:33

hosted version of Bitwarden. I believe it's

51:35

a Rust app. You can

51:37

absolutely run this thing locally. And I'm sure

51:39

that many of us listening have the skill

51:41

set to do it. But it's

51:44

that what happens when all systems are

51:46

down moment and I need that password

51:48

to get back into everything. I

51:51

mean, particularly with Bitwarden only being $10 a

51:53

year for the hosted service, that's a

51:56

value trade off I'm willing to make. That's

51:58

where I sit with it mostly. as well. I

52:00

do think I need to probably develop a

52:02

better practice about backing up and exporting my

52:05

database and then maybe having

52:07

a vault warden instance somewhere on my tailnet that I

52:09

restore into from time to time. You and me both,

52:11

I think. Yeah, so I'll give some

52:13

thought to that. I think it's an interesting idea.

52:16

I wonder if you could do it the other way

52:19

around. If you could treat the vault warden as the

52:21

primary and then bit warden as the cold storage, so

52:23

to speak. Maybe. I'd be down

52:25

for trying that. That does make it easier with everything just being

52:28

on tail scale. There's always a

52:30

lot of layers to think of there. But thank you, Watsy.

52:32

I appreciate that. mascotNR comes in

52:34

with 2000 sats. The last show is my

52:37

first boost, so let's make it two. The

52:39

NR at the end of my name is a

52:42

UK post code, which Alex may recognize as being

52:44

from Norfolk. You

52:46

recognize that from being from Norfolk there, Alex? Yes,

52:48

I do. The post codes

52:51

over there, NR16 and

52:53

NR. I think the one I went and lived

52:55

in Norwich was NR2. Was it

52:57

NR4? Something like that. Anyway, yes, I do. I

52:59

love Norfolk. That's where I did my masters at

53:01

UEA and my

53:03

wife's from Norwich. Very big soft spot

53:06

for that part of the world. Oh,

53:08

he writes, so a JB meetup in Norwich is something I'm

53:10

hoping for. And we'll keep an eye on the meetup page.

53:12

Also Unraid 100% gave me everything I needed for now

53:15

and with my self hosted journey. So I'm pleased they're

53:17

a sponsor. Although I said my last boost, I listen

53:19

to all the shows. I only understand about 1% of

53:22

them. It's that upbeat attitude I'm here for.

53:26

Yeah, well, I was talking to, I think Joe

53:28

over on Late Night Enix was talking about doing

53:30

a meetup in June, which happens to be the

53:32

sort of time I'm going to be in England.

53:34

And I was talking with Wimpy and Popey as

53:37

well about them coming along and trying to

53:39

do something. I don't know if we'll manage to make

53:41

the schedules all work with everybody, but

53:43

I'm considering doing something in Norwich for the

53:45

folks out East or Cambridge area, because

53:48

there's a few folks out that way that I'm talked

53:50

to quite often. And maybe something

53:52

in the sort of Reading, Basingstoke sort

53:54

of area rather than London itself, because I know

53:57

that we did it in London proper last time.

54:00

Depends on where we can find events to do

54:02

it like it could be small low-key just find a

54:04

Riverside pub in fact in Norwich, I know the perfect

54:06

one. So we'll probably do one in Norwich and

54:09

then in the basingstoke area

54:11

Oh, I don't know. We'll find something keep an

54:13

eye on the page. There you go MSC

54:15

0 1 3 5 comes in with our last

54:17

boost that's gonna make it in the show this week. It's 2560

54:20

sats they write I host some things

54:22

on a raspberry 5 in my apartment and some

54:24

services need to be accessible from the internet

54:26

like Matrix for Federation home assistant for the

54:30

Echo in a Google integration. I

54:32

use Cloudflare tunnel in front of caddy since I

54:34

can't forward ports on my apartments network This

54:37

means Cloudflare handles the TLS and can read

54:39

my traffic And grok has this

54:41

issue to and tail scale serve wouldn't let me

54:43

serve my own domain Should I

54:45

worry about my traffic being read by the NSA?

54:47

If so, what should I do? Yes, but

54:50

probably not To the

54:52

point of you know, I don't

54:54

know like tin foil having this particular

54:56

solution There's probably other parts of your life. The

54:58

NSA are much more interested in. Yes, like your

55:00

credit card your debit card Yeah, whatever

55:03

telemetry your car reports those things they love your

55:06

phone account Your

55:08

service provider selling them everything but also if

55:10

you're using matrix that could be SSL already

55:12

in their home assistant as well What

55:15

you could do this is something I've sort of

55:17

been experimenting about making a video for for work

55:19

is Run tail

55:21

scale on a VPS somewhere and run

55:24

tail scale inside your LAN connect

55:27

the two together over the tail net and You

55:30

run the reverse proxy on the

55:32

cloud VPS and then that

55:34

way you can publicly expose that to the

55:36

public internet like you're doing The cloudflare tunnels,

55:38

but you control the TLS you control the

55:40

DNS like you're in charge

55:43

of everything including Making sure

55:45

it's secure because it's on the public internet too, of

55:47

course So with cloudflare you

55:49

do get nice stuff like DDoS protection and all

55:51

the rest of it But if you want to

55:54

own the chain more run

55:56

out on a VPS, thank you very much for the boost ends

55:58

Good question. Let us know what you think you decide to do, MSE?

56:01

Yeah, I'd love to know. We had 10 total

56:03

boosters, so that means there is a couple of boosts in the boost

56:05

bar that we'll have linked in the show notes. And

56:07

we stacked 166,831 SATs with no middleman, nobody in between, just

56:09

an open source peer-to-peer protocol

56:15

sent directly to us to support the show. If you'd like

56:17

to send a boost with a message and some value, get

56:19

a new podcast app at newpodcastapps.com.

56:23

So many great ones to choose from, GPL,

56:25

different models for each type, each individual, each type

56:27

of person. And distros back in the day, where

56:30

each one really stood out as having its own

56:32

unique proposition, that's where we're at with the

56:34

2.0 apps. Check it out at newpodcastapps.com.

56:37

And don't forget selfhosted.show slash SRE, if

56:39

you would prefer to pay in fiat

56:41

fund coupons. You can support the

56:43

show over there when you get an ad-free feed along

56:45

with our post show. Yeah, a little extra content to

56:48

say thank you for our members. We

56:50

really appreciate you. You can find that at

56:52

the website. There's a website,

56:54

selfhosted.show. You know about

56:57

that? It's got links. It's a

56:59

pretty cool website, huh? Yeah, easy. Like

57:01

one spot, you can just link to stuff. Now,

57:04

if you'd like to come see your

57:06

favourite fat bearded Englishman in England, meetup.com/Jupiter

57:09

Broadcasting, keep an eye on that page

57:11

for all the info regarding meetups. You

57:14

can also find me on the internet at

57:16

alex.ktsz.me. You can

57:18

find me at chrislas.com. And until

57:20

next time, thanks for listening. That was selfhosted.show

57:22

slash 123.