0:00

This week on the show

0:02

we are looking at key

0:04

considerations for benchmarking network storage

0:06

performance. Open ZFFs is 2.3.0

0:08

availability and the updates on

0:10

Asia BSD Con's conference and

0:13

the upcoming ghost BSD Desktop

0:15

Conference as well. Recovery from

0:17

an external ZZ route is

0:19

our very own Tom Jones

0:21

teaching us. how to create

0:23

a new issue in a

0:26

guitar repository with Ansible stories

0:28

that UNIX refuses to believe

0:30

the date limits in

0:32

UFS-1 having recently got

0:34

an extension and more in

0:36

this week's episode of BSD. Hey,

0:53

how are things on your end?

0:55

Hey, hey, hey. Well, my end,

0:57

it's been hot here. It's been

1:00

39 degrees for the last three

1:02

days. So that's 39 degrees Celsius

1:04

for you, people over in the

1:07

US of A. But everywhere else

1:09

in the world that follows metric,

1:11

it was 39 degrees Celsius for

1:13

last three days. So it's actually

1:16

been very hot. Had to sleep

1:18

with the air condition on last

1:20

night. It didn't get below. What,

1:23

26 overnight? Oh yeah. So yeah, it's, it's been

1:25

a scorcher of a summer over here,

1:27

we're still dealing with fires down in

1:29

Victoria, so. Yeah, that's the drawback of

1:31

that, that there's a lot of fire.

1:33

Because what comes with this is we

1:36

get changes and we'll get a storm

1:38

front that rolls through that's full of

1:40

lightning and that starts more fires again.

1:42

Yeah. And there's no rain. It's like

1:44

the rain doesn't even hit the grounds.

1:46

So it's... And you have to sleep

1:48

in the fridge basically too. Yeah, yeah,

1:50

it's like, you know, last night I

1:53

put the air conditioning, the other two

1:55

nights we sort of, we put up with

1:57

not having the air conditioner on, but

1:59

I said... need a decent night's sleep

2:01

so the air conditioner and the ceiling

2:03

pan went on to keep cool. Yeah,

2:05

if you can't do sleep that's affecting

2:07

you for the rest of the day

2:09

and it's difficult to get it. I

2:11

did notice I was a lot sharper

2:13

today, so that is a bonus that

2:15

shows you that yes, the 1.2 kilowatts

2:17

of air conditioning power that was being

2:20

consumed every hour overnight was worth it.

2:22

Yeah I mean you collect also a

2:24

lot of sun because a lot of

2:26

sun is actually... Yeah well we get

2:28

we get a lot of sun but

2:30

unfortunately I don't have batteries so you

2:32

know you can't use that overnight

2:34

no I've got an EV that

2:36

that's that's that's full can't charge

2:38

anymore in that but yeah so

2:40

that's that's that's been the general

2:42

general weather here so what are

2:44

you being up to Benedict? Busy,

2:47

I mean this semester is

2:49

coming to an end and

2:52

the exams are starting next

2:54

week so I'm preparing

2:56

crazy questions for students

2:58

to solve and exercises

3:01

for the exam so that's

3:04

currently on my plate. Other

3:06

than that, yeah so far so good

3:08

I have a couple of things

3:11

in the background going on

3:13

at the department itself. But

3:16

I could probably talk about

3:18

that later when it actually

3:21

materialized when there's something concrete

3:23

to talk about. I've been watching the

3:25

BSD space recently, some messages going

3:27

by and we'll cover that in

3:29

the show later today, but yeah,

3:31

so far so good. Excellent. Well,

3:34

there's been a fair bit happening

3:36

just in a general sense. You

3:38

know, this is this is for

3:40

you patron, patron listeners out there.

3:42

So... you know come around 600

3:44

we're just trying some things out

3:46

so you guys are the only

3:48

ones that are going to get

3:50

this conversation this will be all

3:52

edited out of the main show

3:55

it's it's still tech-related but there's

3:57

no articles to go with it so

3:59

this is more of our observation

4:01

that's gone on across the the

4:03

fatty verse because that's where most

4:05

of us hang out I'm still

4:07

trying to convince Benedict to come

4:09

and join us in the fatty

4:12

verse that's what this is about

4:14

okay yeah it's the shame you

4:16

in the come and join us

4:18

now we've all we've all left

4:20

X you know that is a

4:22

thing of the past Yeah, our

4:24

accounts are dead and buried and

4:26

there's like so many of us

4:29

have just torched our accounts and

4:31

just like if you see our

4:33

previous handles that's not us anymore.

4:35

We deleted the accounts and just

4:37

wet. It doesn't even have a

4:39

real underscore Benedict or Jason under

4:41

before it. Yeah, yeah, the flaming

4:44

fire that is X. But yeah,

4:46

over on the Master Don that's

4:48

been going on. So our friend

4:50

Stefano, we, you know, feature his.

4:52

you know, articles quite frequently on

4:54

the show and also we've had

4:56

an interview recently with him. He

4:58

started up this service called Fedimetto.

5:01

So that's FEDI-M-E-T-E-O. So what that

5:03

is is basically a feed from

5:05

all the bureau meteorology around the

5:07

world. So he's capturing that and

5:09

then feeding it using, I think,

5:11

snack too, which is the mastodon

5:13

server. and feeding that into people's

5:16

feed. So basically you can subscribe,

5:18

like follow different things like I

5:20

follow at Melbourne and at Wodonga

5:22

at Fedimeto and that then means

5:24

that anything that comes and updates

5:26

on those accounts will actually fall

5:28

into my feed. So you know

5:30

I get frequent weather updates and

5:33

that sort of stuff that fall

5:35

in my feed. So it's quite

5:37

a quite a good service that

5:39

can give you what the current

5:41

temperature is, forecast and things like

5:43

that. People say, you know, the

5:45

Fediverse is not exactly welcoming and

5:47

it's really what you make it.

5:50

So think of the old Twitter

5:52

where you actually... had a buy

5:54

a hose of information and it

5:56

was all date-based. So you could

5:58

basically go through, you wouldn't miss

6:00

anything. Now it's all algorithm-based and

6:02

driven like that. So, you know,

6:05

it's whatever the service like, you

6:07

know, X or Meta or whatever

6:09

want to pump in front of

6:11

your eyeballs. Basically, that doesn't happen

6:13

on the Fediverse. And people go,

6:15

oh, there's nothing in my feed.

6:17

Well, you've got to make that

6:19

happen in your feed. So, fortunately,

6:22

before all the APIs on X

6:24

got burnt to hell, you could

6:26

basically extract your follow-a-list out and

6:28

then it would do a search

6:30

across the Maston world to actually

6:32

go and find those particular users

6:34

that you're following on Twitter and

6:36

then... then get allow you to

6:39

follow them on must it on

6:41

so you know I had quite

6:43

a collection that came across which

6:45

was really good because then that

6:47

meant that my feed was populated

6:49

with people that I'd follow and

6:51

as it got Twitter got worse

6:54

and worse more people came over

6:56

people would like retweet or it's

6:58

retoot in master must it on

7:00

terminology and eventually you know you

7:02

come across all the people that

7:04

you were following and then you

7:06

build your community up and so

7:08

then your feed becomes so full

7:11

that overnight I get through about

7:13

you know the last five hours

7:15

and I miss so much that

7:17

happens on must on these days

7:19

that I just can't keep up

7:21

but it's good because it means

7:23

that you know you can have

7:25

conversations that sort of stuff and

7:28

the trolls don't find you because

7:30

it's a very curated event feed

7:32

and you only sees things you

7:34

want to see and not something

7:36

that people push into your timeline.

7:38

Correct. And it's like so people

7:40

have to go looking for you

7:43

rather than you know, oh let's

7:45

let's cause some friction here, it

7:47

gets more interaction. and keeps people

7:49

engaged on the platform, that's not

7:51

how it works. So it means

7:53

that you can end up floating

7:55

around in the universe and not

7:57

actually come across the trolls and

8:00

things like that. So it's quite

8:02

good that way and it's got

8:04

tools that you can block trolls

8:06

on things if they do come

8:08

across you, but I've never had

8:10

to do that. So yeah, back

8:12

on the Fetti Meadow, it's really

8:15

good service. So if you're on

8:17

Master Don, go to Feddy Meadow,

8:19

hang on, like, because we're not

8:21

shooting a shooting a show, shooting

8:23

a show, we can actually, we

8:25

can actually, take time delay here

8:27

and not actually cause our editor

8:29

or any any particular issues. So

8:32

the actual service that I'm following

8:34

is a you dot Fedimeto.com. I

8:36

think there's a EU one and

8:38

other ones around the world. So

8:40

if you go and check Fedimeto.com,

8:42

F-E-E-D-I-M-E-T-E-O, dot com. that should take

8:44

you to a link page that

8:46

will take you further on. So,

8:49

and then you can just follow

8:51

all the accounts that the places

8:53

you visit pretty frequently. So, it

8:55

would be ideal for people that

8:57

they're going to BSD conferences to

8:59

be able to find out what

9:01

the weather is at those places

9:04

and be able to get a

9:06

feel for it leading up to

9:08

their travel to those places. So,

9:10

it's got its uses for us

9:12

and the rest of the community.

9:14

Oh yeah, and it's great that

9:16

people build these kinds of services

9:18

and I guess there's plenty of

9:21

BSD in the background doing all

9:23

that work. Oh yeah, he's got

9:25

his jails pumped. Yeah, so I

9:27

think he was looking at a

9:29

coma server instance, but I believe

9:31

for all these he's using snack

9:33

too because it's simplistic, it doesn't

9:35

need to be have all the

9:38

tools that you get with a

9:40

fully fledged mastodon or a coma

9:42

or miskey instance and that sort

9:44

of stuff. Yeah, he's kept it

9:46

simple, but it's all running in

9:48

free BSD jails and things like

9:50

that. So I think each stack

9:53

instance has its own jail to

9:55

be able to keep them all.

9:57

isolated, which is pretty cool. Yeah,

9:59

that makes sense. And it's another

10:01

way of eating your own dog

10:03

food, like we talk a lot

10:05

about BSD and he certainly has

10:07

done so. And it's kind of

10:10

nice to actually build these things

10:12

and see, hey, this is a

10:14

viable way of doing this without

10:16

the usual systems that are typically

10:18

thrown around. So it's all easy,

10:20

fairly easy, I guess, and doable

10:22

with the building blocks that the

10:24

BSDs provides. Yeah. So yeah, we

10:27

thought we'd share that with you

10:29

before we started the main show.

10:31

And yeah, if you've got feedback

10:33

about the pre-show, you know, we're

10:35

quite happy to take that feedback

10:37

on board. And if there's things

10:39

that you want us to discuss

10:42

that maybe not quite right for

10:44

the show, but you think that,

10:46

you know, you'd like to hear

10:48

what our thoughts are that are

10:50

not. on the full public record,

10:52

but on the sort of a

10:54

bit more of a private record.

10:56

We're quite happy to do that

10:59

too. So send your feedback into

11:01

Feedback at BSD. Now. TV. Let

11:03

us know that you're a Patreon

11:05

supporter. So then that way we

11:07

know how to answer and where

11:09

to answer that. So we can

11:11

have like effectively two feedback sections.

11:14

You can have the pre-show feedback.

11:16

So if you're a Patreon supporter,

11:18

let us know. in your feedback

11:20

that you are a supporter so

11:22

that way we can work out

11:24

where to put your question and

11:26

answer. Is it? All right then

11:28

let's start the official show. Right,

11:31

yeah, is this where we're going

11:33

to be quiet now Benedict? They're

11:35

going to know the secret source

11:37

now because we'll be quiet and

11:39

then it'll be edited magically and

11:41

then all of a sudden, oh

11:43

wow, this is when the show

11:45

starts. It's like I was behind

11:48

the scenes when it happened. Yeah.

11:50

This is now episode 598 UFS1

11:52

up to date. Recorded on the

11:54

4th of February 2025. Ha! I

11:56

got the year correct. This episode

11:58

of Bisti now is brought to

12:00

you by Tar Snap. Go to

12:03

Tar Step.com/Bisti now to find online

12:05

backup for truly paranoid people. And

12:07

if you want to support this

12:09

show in one way or the

12:11

other, check out our Patreon page

12:13

at patreon.com/Bisti now. And we thank

12:15

you in advance for that. Hi,

12:17

I'm your host Benedict Rachling. And

12:20

I'm Jason Tubner. Hi. Can you believe

12:22

that a one month has already passed?

12:24

It's like over so quickly and

12:26

we're in February. Yeah, where's the

12:28

year going Benedict? It only feels

12:30

like I've still got stuff here

12:33

that I haven't unpacked from EuroBSTCon

12:35

right? And that was how long

12:37

ago now? I can't believe how

12:39

much time has passed. I'm probably wearing

12:41

my speaker gift, the scarf we got.

12:43

So that is certainly looking nice. You

12:45

have to wait a little... I'm not

12:48

sure your weather is scarf worthy. No,

12:50

not yet. but yeah we had a

12:52

bit of a pre-show discussion about some

12:54

of the stuff that's been going on

12:56

in the community. So we talked about

12:58

what it's like down here at the

13:00

moment. So you know it might convince

13:02

a few more of you to become

13:04

patrons for the show. There's more detail

13:06

we've got in the show notes and

13:08

we'll talk about it later on in

13:11

the show but yeah we had a

13:13

bit of a pre-show discussion about some

13:15

of the stuff that's been going on

13:17

in the community and what we've been

13:19

up to and what it's been like

13:21

for us for us around around. now

13:23

where we live so yeah general general

13:25

chit chat so patreon.com come and join

13:27

us and you'll be entitled to the

13:30

pre-show too. All right what is

13:32

available for everyone is

13:34

of course our headlines

13:36

this week as always

13:39

and it starts with

13:41

another good article by

13:43

Clara Systems considerations or

13:45

key considerations even for

13:47

benchmarking network storage performance

13:50

written by None other

13:52

than our producer

13:55

JT. So this is

13:58

about network. storage

14:00

performance, so a lot of

14:02

things in there, and it

14:04

introduces with when measuring network

14:06

storage performance, it's essential to

14:08

account for factors like protocols,

14:10

workloads and testing tools. SMB,

14:12

NFS and IceGazie each bring

14:15

unique strengths and challenges to

14:17

the table, learn how to

14:19

optimize your NAS setup for

14:21

accurate and meaningful benchmarking results.

14:23

So these are the three

14:25

parts, SMB, server message block.

14:27

aka Windows file sharing, NFS

14:29

network file system, the Unix

14:31

file sharing system and ISKASI

14:33

for exporting storage and importing

14:35

it as a local storage

14:38

even though it's coming over

14:40

the network. The

14:42

article goes. When setting up a

14:44

network attached storage, or a NAS

14:46

for short, system using ZFS, it's

14:48

essential to understand the performance users

14:50

can expect when accessing files. And

14:53

NAS on the office network, on

14:55

a VPN, or in the cloud,

14:57

acts as a central hub for

14:59

data. Its ability to handle multiple

15:01

users reading and writing files efficiently

15:03

is critical to its success. Performance

15:05

testing is a valuable step to

15:07

gorge how well the NAS will

15:10

serve its users. but it's not

15:12

always a straightforward task. Different protocols,

15:14

SMB, NFS and ISKASI handle file

15:16

operations in unique ways, resulting in

15:18

their own unique challenges. Workloads vary

15:20

significantly and each workload is different.

15:22

An application that accesses large volumes

15:24

of small files will need different

15:26

tuning than video editing software and

15:29

is streaming a small volume of

15:31

a very large file. And

15:33

each different protocol has its benefits.

15:35

For instance, SMB is optimal for

15:37

compatibility across operating systems. NFS for

15:40

rock solid stability for Unix systems

15:42

and IceGazie provides block level access

15:44

for virtual machines. However, they also

15:46

have unique details that need to

15:48

be considered when seeking to maximize

15:51

their performance. Before diving into testing,

15:53

it's important to consider these factors

15:55

to ensure a meaningful and accurate

15:57

assessment of your analysis capabilities. So

15:59

they start with benchmarking. S&B over

16:01

the network. And what are the

16:04

challenges here? Using ZFS as the

16:06

NAS for Windows clients or mixed

16:08

client infrastructures typically accomplish through the

16:10

use of S&B shares. S&B or

16:12

the server message block is a

16:15

network file sharing protocol that allows

16:17

applications and uses to access files

16:19

and other resources on a network

16:21

computer. Originally developed by IBM and

16:23

further enhanced by Microsoft S&B is

16:26

widely used in Windows environments. It's

16:28

compatible with other operating systems. SMB

16:30

enables users to interact with files

16:32

and directories on a remote server

16:34

as if they were on their

16:36

local machine. Its versatility makes it

16:39

popular choice or a popular choice

16:41

for file sharing on home business

16:43

networks. Crystal disk mark is a

16:45

popular Windows IO testing tool and

16:47

can be attempting method to test

16:50

SMB shares. The tool has some

16:52

configuration options that can be adjusted

16:54

to influence how the test is

16:56

run. However... It was designed for

16:58

testing local disks and its lack

17:01

of precise control over how the

17:03

IOSA issued results in varied reliability

17:05

over the network. It's a great

17:07

utility for gaining a quick estimate

17:09

of a system's potential performance, but

17:12

it should not be even identified

17:14

for high accuracy results. Okay, so

17:16

that's good to know, and they

17:18

reference also the CRISC diskmark FAQ

17:20

for more details. What are other

17:22

tools that are in that area?

17:25

There is black magic design raw

17:27

speed test. The BMD RASB test

17:29

is especially useful for simulating a

17:31

media- heavy workload, yet it would

17:33

not be as reliable for testing

17:36

other workloads. Then there is the

17:38

industry standard file for testing SMB

17:40

that they also recommend, and they

17:42

have covered in other articles. However,

17:44

there are some special considerations when

17:47

running file on Windows systems. Because

17:49

of the caching Windows is doing,

17:51

if you're testing using PowerShell on

17:53

a Windows box to an SMB

17:55

share, you will need to use

17:57

the dash- dash direct flag. This

18:00

avoids interference from the Windows IO

18:02

Cashing subsystem. And Clara would recommend

18:04

you avoid attempting to run file

18:06

in a... WSL environment, the mix

18:08

of emulation, virtualization, and cashing within

18:11

WSL can produce very under-reliable results.

18:13

The key metrics here are Iops,

18:15

latency throughput from the benchmark, CP

18:17

usage by the S&P client, and

18:19

the OS network stack on the

18:22

device under test as well as

18:24

on the server. Then they talk

18:26

a little bit about optimizing performance

18:28

with SMB multi-channel support. Deployments of

18:30

an SMB server and client can

18:33

vary significantly based on the network's

18:35

capabilities and the type of workloads.

18:37

Enabling multi-channel support may lead to

18:39

much greater performance in these scenarios.

18:41

First, this will have to have

18:43

the client make multiple TCP connections

18:46

to the server, allowing the transfer

18:48

to be spread over multiple control.

18:50

and using multiple network interfaces concurrently,

18:52

as well as allowing the SMB

18:54

server to use multiple threats to

18:57

parallelize I.O. operations. And if the

18:59

network is on a LAGG, link

19:01

aggregation, i triple E, A02.3 AD

19:03

for the nerds, multiple flows can

19:05

also allow the load to balance

19:08

across multiple links and switch ports

19:10

offering greater bandwidth. Okay, let's switch

19:12

over to NFS benchmarking. NFS and

19:14

ZFS block size is important here

19:16

when using NFS to share ZFS.

19:18

It's important to ensure that the

19:21

NFS block size and ZFS record

19:23

size are matched. If they are

19:25

mismatched, there can be dramatic results

19:27

in performance and need to a

19:29

lot of penalties that people experience

19:32

when accessing and writing data to

19:34

it. There is a separate article.

19:36

They reference here as ZFS performance

19:38

optimizing success stories. that I think

19:40

we also covered in previous episodes.

19:43

So this is good to have

19:45

read again if you are trying

19:47

to implement this. The two primary

19:49

performance issues that can occur are

19:51

read in right amplification and inflation.

19:53

What's this about right amplification? occurs

19:56

when the physical right-to-disk must be

19:58

larger than the logical right, the

20:00

data that is changing. With the

20:02

default ZFS configuration of 128 kilobite

20:04

records, any changes to files smaller

20:07

than the record size, such as

20:09

32 kilobites, require rewriting the entire

20:11

128 kilobite record. This occurs even

20:13

though only 32 kilobites of data

20:15

have been modified. As a result,

20:18

the system uses four times the

20:20

disk bandwidth compared to a scenario

20:22

where the record size matches the

20:24

size of the changes. Same or

20:26

a similar issue can occur during

20:29

reads if the record sizes one

20:31

megabyte. An application requires only four

20:33

tiny kilobites of data. ZFS must

20:35

read the entire one megabyte record,

20:37

verify the check sum, and then

20:39

return just a tiny tiny four

20:42

kilobites the application needs. Subsequent requests

20:44

for the next four kilobites will

20:46

typically be served from the ZFS

20:48

arc, making the impact less significant

20:50

as the data was prefached. However,

20:53

The application accesses a random subset

20:55

of blocks that if this might

20:57

end up reading 256 times the

20:59

amount of data actually required to

21:01

fulfill those requests. Not so good.

21:04

What's right inflation? Right inflation is

21:06

when a single block of a

21:08

file is updated. The manifest must

21:10

update the indirect block that points

21:12

to it due to copyright. It

21:14

also updates the parent blocks and

21:17

the parent blocks of those parents

21:19

all the way up to the

21:21

route to the tree. For each

21:23

single block of change, manifest updates

21:25

metadata blocks that track it, sometimes

21:28

multiplying or multiple times for data

21:30

redundancy. So a 32 kilowatt change

21:32

could trigger 3 to 5 additional

21:34

128 kilowatt rights, especially if changes

21:36

happen far apart within the file.

21:39

The disinflation decreases disk usage and

21:41

slows down performance, particularly when changes

21:43

are scattered across the disk. And

21:46

they talk about configuring NFS rights

21:48

for better performance. There are a

21:50

few other details to keep in

21:52

mind when testing an NFS share.

21:54

For example, the default number of

21:56

the NFS strikes can vary across

21:58

Unix operating system on net BSD.

22:00

is four threats. Three BSD has

22:02

eight threats per core and on

22:04

the OS, the default is 16.

22:07

Lastly, most Linux distros have a

22:09

default of eight threats. Okay, that's

22:11

important to distinguish. And to talk a little

22:13

bit more about key differences between

22:15

NFS three and four, they also

22:18

have a separate article on PFS,

22:20

the parallel NFS implementation on free

22:22

BSD. They also link to that

22:25

article. They covered a couple weeks

22:27

ago. Then there is benchmarking ice gazzy

22:29

over a network. Similar to NFS it's

22:31

important to ensure that the ice gazzy

22:33

block size and set of its record

22:35

size are matched. The same read and

22:37

write amplification and inflation performance issues can

22:40

occur. These are mismatched and with ice

22:42

gazzy situation can be much worse as

22:44

typically the ice gazzy block size will

22:46

be 512 bites or four kilabytes, a

22:48

sector size. There are a

22:50

few other minor considerations to keep

22:53

in mind. For example, jumbo frames.

22:55

While critical to performance in the

22:57

past and still helpful in very

22:59

specific situations, these days are often

23:01

more trouble than they are worth.

23:03

Modern Knicks using TSO, the TCP

23:05

segmentation of loading, which passes data

23:08

to the Nick in 64 kilowatt

23:10

chunks, ice-gazi specific offload features and

23:12

hardware acceleration, do not require reconfiguration.

23:14

Your entire network to get the

23:16

desired performance. So they talk

23:18

a little bit about each of these. If you want

23:20

to check out the full article, but

23:23

they conclude the whole thing with final

23:25

considerations for optimizing network storage.

23:27

Most good storage benchmarks of the

23:29

network start by determining the performance

23:31

of the storage without the network

23:33

directly to the server. Baseline value

23:36

that this creates helps determine how

23:38

much performance is being lost to

23:40

the network. It also identifies configuration

23:42

or topology changes that could reduce

23:44

the network's impact on storage workloads.

23:47

Most of them, that one of the most important

23:49

factors to consider when benchmarking of the

23:51

network is the additional latency. Packets having

23:53

the traversed the network will increase both

23:56

the amount of time and the variance

23:58

in the time, the jitter. Some

24:00

of this can be overcome with

24:02

additional concurrency to keep more data

24:04

and flight at once, yet depending

24:07

on the workload that may not

24:09

be possible. If you are reading

24:11

a database knowing which rows you

24:13

need to read next will not

24:16

be possible without the result of

24:18

the current read operation. And with

24:20

the right design, planning and tuning,

24:22

the network can be made to

24:25

achieve your performance targets. As with

24:27

all things, it's just a series

24:29

of tradeoffs between throughput, latency, complexity

24:31

and cost. If you are looking

24:34

to design your next search and

24:36

network infrastructure, consider Clara Storage Design

24:38

and Implementation Solution. That's a good

24:40

article. There's a couple of things

24:43

I sort of want to cover

24:45

before we move on the NTU

24:47

section that they talked about, which

24:49

was labelled Jumbo frames. So that's

24:51

correct. You don't really want to

24:54

have your ice scuzzy traffic anyway

24:56

moving across segments from layer two,

24:58

layer three. So you really want

25:00

to be keeping your ice scuzzy

25:03

target and your initiator on the

25:05

same segment. It can be a

25:07

V land or a separate isolated

25:09

switch, but you want to keep

25:12

that in the layer two segment.

25:14

The TSO will... give you the

25:16

performance that you need without having

25:18

to do empty you adjustments. And

25:21

I have seen points where if

25:23

you traverse across too much of

25:25

a network anyway, your empty you

25:27

can be fiddled with depending on

25:30

the layer of network and the

25:32

control if it's in your control

25:34

or how do you control. So

25:36

yeah, keep it layer two. Don't

25:39

worry about using jumbo frames and...

25:41

utilize TSO which came around I

25:43

think free BSD 12 we saw

25:45

the implementation that we didn't have

25:48

to worry about three BSD 11

25:50

this is I come across a

25:52

heap of TSL issues when I

25:54

went from 11 to 12 for

25:57

beehive so well versed in that.

25:59

So there's also another one I

26:01

wrote an article for the BSD

26:03

journal around IceGuzzy a while back

26:06

so you might want to go

26:08

and check that out because there

26:10

is actually some design or configuration

26:12

considerations you need to take into

26:15

account when you actually are using

26:17

ZFS backed by Scuzzi storage and

26:19

how it presents itself to an

26:21

initiator. So go check my article

26:23

out to plug for myself. And

26:26

yeah and go and check those

26:28

because actually made myself go find

26:30

those those sharp edges so you

26:32

didn't have to and have to

26:35

scratch your head wondering what's going

26:37

on. And you know NFS that's

26:39

another one it's like NFS is

26:41

one of those ones where it

26:44

really needs to die but it

26:46

won't die and things you need

26:48

to consider is like you know

26:50

a lot of clients as it

26:53

talks about in the article are

26:55

NFS 3. So you know take

26:57

that what you will. NFS4 is

26:59

good because you've got more ACLs

27:02

and that sort of stuff, but

27:04

overall NFS as we've talked about

27:06

and feedback in the past doesn't

27:08

support file locking. So depending on

27:11

your application and what your use

27:13

case is, NFS might not be

27:15

the fit for you and that's

27:17

where IceGuzzi would probably be a

27:20

better choice in that situation. So

27:22

other than that there's some really

27:24

good stuff in this article. Oh

27:26

yeah, I forgot about that. Multichannel.

27:29

support in SMB. So that's a,

27:31

yeah, that's, so on a typical

27:33

single thread, if you don't tune

27:35

your SMB protocol, your single threaded

27:38

will basically top out probably around

27:40

about 15 gig mark, 15 gigabits

27:42

per second, not gigabits per second,

27:44

gigabits per second, gigabits per second.

27:47

So, you know, on a single

27:49

thread on a 10 gig on

27:51

modern hardware, you're pretty well okay,

27:53

but this is where, you know,

27:55

you want to have multi-threaded support,

27:58

so you can use LACP or

28:00

lagging. that you might have for

28:02

dual connection to your storage network.

28:04

So yeah, that's a very good

28:07

feature. you nice to have and

28:09

configure it up it's there it's

28:11

available so use it so good

28:13

good work there JT thumbs up

28:16

mate yeah good good and we're

28:18

staying a little bit in the

28:20

storage area yes staying in the

28:22

storage area and still in the

28:25

headlines as well I'd like to

28:27

say so the release of ZFS

28:29

2.3. 0 so there was a

28:31

release notes that came out from

28:34

the Open ZFS project and they

28:36

are excited to announce release and

28:38

you know we're all excited that

28:40

it's come about because there's some

28:43

really really good features in this.

28:45

A lot of stuff, a lot

28:47

of things that people have been

28:49

waiting for years for and it's

28:52

finally now in a production state

28:54

where we can all use it.

28:56

So key features in Open ZFS

28:58

2.3. Red Z expansion, adds a

29:01

new device as to an existing

29:03

Red Z pool, increasing storage capacity

29:05

without downtime. So this is where,

29:07

you know, we have always long

29:10

for the ability to go, oh,

29:12

my Z pool is, oh, my,

29:14

my, uh, pool is basically full

29:16

and I need to add more

29:18

storage to it. And it really

29:21

basically required you to ZFS and

29:23

all your data sets across to

29:25

something else. Pull down the pool,

29:27

recreate the pool and then send

29:30

them all back again. Now with

29:32

the Rades E expansion means you

29:34

don't have to go through that

29:36

particular process. So it's good to

29:39

have. Remember there's going to be

29:41

some performance issues until the rewriting

29:43

of data has gone on on

29:45

the whole. pull itself as it

29:48

balances itself out but it means

29:50

it's a lot less work for

29:52

administrators especially when you've got say

29:54

100 terabytes that you need to

29:57

add more storage too it's it's

29:59

a really good feature to have

30:01

and for Home Lab it's probably

30:03

the number one feature that everybody's

30:06

been looking for too. Yeah because

30:08

it gives people option to migrate

30:10

slowly to a new array type

30:12

without you know backing up destroying

30:14

and recreating the whole thing. Yeah I've

30:16

had to do it many times with

30:18

with different things here and even in

30:21

a business setting I've had to do

30:23

it so this is a really good

30:25

feature that's going to save a lot

30:27

of people a lot of time. Also,

30:30

we see as key features is our

30:32

fast-de-dupe. It's a major

30:34

performance upgrade to

30:37

the original opens-of-s-duplication

30:39

functionality. So, de-duplication

30:41

is got its

30:44

benefits, but the original

30:46

implementation was

30:48

actually quite poor. It

30:51

actually went against you

30:53

anyway. So, I think Jim Salter

30:55

and gave a bit of a talk.

30:57

on this on one of the Clara

30:59

webinars that was with

31:01

Alan and Jim at not

31:04

that long ago actually. So

31:06

you know go to their

31:08

webinar page on the Clara

31:10

systems website and they go

31:12

through this particular functionality so

31:14

then you can get a

31:16

bit more of an insight

31:18

into how the fast G-Dup

31:20

is different from the

31:22

original D-Dup. and some considerations you

31:24

need to take on board when

31:26

you're using any sort of deduce

31:28

with set of tests. So that's

31:31

a good feature too. We also see direct

31:33

IO which allows the bypassing of the

31:35

arc for reads and rights, improving

31:37

performance in scenarios like envy me

31:39

devices where casing may hinder the

31:41

efficiency. So we know that envy

31:43

me is right on the PCIE

31:45

bus for super fast access. sometimes

31:47

Ark read and write will introduce

31:50

latency into that because you've got

31:52

storage that's basically working nearly at

31:54

the speed of the RAM, but

31:56

if you've got to do multiple

31:58

transactions and push data to to

32:00

RAM and read it from RAM, then

32:02

you've got that extra latency instead of

32:04

just committing it straight to storage that

32:06

can handle the performance that the CPU

32:09

is looking for, can actually hinder that.

32:11

So it's good to see that is

32:13

also a feature. Jason output for most

32:15

used commands. So this allows you to

32:17

generate machine readable output from... certain commands

32:19

that you run with ZFS and Zpool

32:22

commands. This will help those that are

32:24

writing scripts that need to pass certain

32:26

output of commands to do automation. So

32:28

Jason gives us the ability to do

32:30

that, whereas the other output that we're

32:32

traditionally used to is human readable output

32:35

button machines can't interpret that or you've

32:37

got to do some huge amounts of

32:39

ork to get the results that you're

32:41

looking for. So Jason just makes it

32:43

so much easier to get that data.

32:45

We see long name support so supports

32:48

the file and directory names up to

32:50

a thousand twenty three characters. Yes I've

32:52

hit this limitation the previous two fifty

32:54

file limitation because people have got Windows

32:56

shares and they want to move that

32:58

to an archive server and the archive

33:01

servers ZFS run in December and they

33:03

try to drag the stuff over and

33:05

bam they hit issues because they've got

33:07

huge long file names, huge long directory

33:09

names. The long file name support on

33:12

ZFS now fixes that particular issue so.

33:14

Again, another thing that I look forward

33:16

to in 2.3. We also see bug

33:18

fixes, a series of critical bug fixes

33:20

addressing issues reported in previous versions, numerous

33:22

performance improvements throughout the code base, and

33:25

see supported platforms now with Linux kernels

33:27

4.18 to 6.12, so that basically covers

33:29

you for... all year Linux enterprise distros

33:31

as well as I think Fedora 41

33:33

is covered in 6.12. So

33:35

that's good to see

33:38

that there's that

33:40

support for all the

33:42

basically main distributions

33:44

and Linux kernels that

33:46

are being used

33:48

in production today. But

33:51

for us, what

33:53

we care about FreeBSD

33:55

has seen 2 .3

33:57

be installed into

33:59

13 .3. So if

34:01

you're at 13 .3,

34:04

which is end of

34:06

life, the 13 .4

34:08

is now out,

34:10

13 .4 is your

34:12

go to. So you've

34:15

already on 2 .3.

34:17

If you've upgraded

34:19

your Z pool, of

34:21

course, and FreeBSD

34:23

versions 4 .14 .0 to

34:25

14 .2 is also

34:28

supporting 2 .3 .0.

34:30

There's further information on

34:32

the release notes.

34:34

So I highly recommend

34:36

you going to

34:38

our show notes and

34:41

clicking the link

34:43

and checking out the

34:45

rest of the

34:47

release notes there. But

34:49

if you go

34:51

over to the OpenZFS

34:54

.org website, you will

34:56

actually see the

34:58

instructions. If you're a

35:00

Linux user, how

35:02

to install it, but

35:05

FreeBSD users, enjoy

35:07

the ride. And

35:09

yeah, if you upgrade your pool,

35:11

don't forget to upgrade your boot blocks.

35:13

So that also the pool gets

35:15

recognized, especially when you boot from that

35:17

pool. Yes, and also do a

35:20

G -part before you go to upgrade

35:22

your boot blocks, because yeah, people get

35:24

caught out all the time when

35:26

they don't realize they're using UEFI and

35:28

they go and blat the boot

35:30

block straight over the petition that's holding

35:32

UEFI. And they just get the

35:34

prompt that they can't find an operating

35:36

system. So yes, make sure you

35:38

do a G -part show to make

35:40

sure that you what type of booting

35:42

you're using and make sure you

35:44

populate the UEFI shim or your boot

35:46

block accordingly. Okay,

35:57

time now for

35:59

the... Newsround up this

36:01

week. Remember last time Jason and

36:04

I were recording and we mentioned

36:06

a bunch of conferences going on

36:08

and we were not sure about

36:10

what Asia BSTCon was about and

36:13

now we have official a word

36:15

from them in their announcement mailing

36:17

list. They said... the following about

36:19

Asia BSD-Con. So a couple of people

36:22

were wondering, is it actually happening? When

36:24

is the call for paper opening and

36:26

things like that? So here they write,

36:28

dear Asia-BSD-Con supporters, we know that many

36:31

people have been wondering about the status

36:33

of this year's Asia-BSD-Con in Tokyo Japan,

36:35

and we apologize for the lateness of

36:38

our communication and the message it must

36:40

convey. Due to a conference of inferences

36:42

or circumstances, financial, logistical and personal,

36:44

we are unable to hold the

36:46

conference as planned this year. We

36:48

deeply apologize for this, Kometasai, for

36:51

this and are sad that we

36:53

will not be able to meet

36:55

in Tokyo this year as we

36:57

had planned. Asia Bisticon has weathered

36:59

financial downturns and earthquake and other

37:01

very real challenges since it was

37:04

started 20 years ago. For a

37:06

conference... to have a vibrant future,

37:08

some things must change and we

37:10

are undertaking these changes presently

37:12

and also going to be

37:14

asking for help from the community.

37:17

So there's no age obesity con

37:19

2025, this is the immediate cancellation

37:22

announcement here. They also

37:24

tell us about creation of

37:26

a transition organizing committee, including

37:29

when Philip Pub's George Neville

37:31

Neal and Sato San, Hiroki

37:33

Sato, and... There is a call

37:35

for recruitment to the 2026 Asia

37:37

BSTCon program committee. George Neville Newell

37:39

has agreed to head that program

37:42

committee and recruit people by volunteering

37:44

now who greatly aid their work

37:46

for the 2026 conference. And they

37:48

also call for financial support of

37:50

the conference. The biggest challenge to

37:52

Asia BSTCon has been inconsistent funding,

37:54

sometimes due to internal issues within

37:56

the project and at times due

37:58

to the vagaries. of global finance

38:01

including fluctuations in the rate of exchange

38:03

between the currencies used. As many of

38:05

you know Asia-based account pays for travel

38:07

and launching for speakers and this is

38:10

a significant part of our costs and

38:12

we do not see that changing. Travel

38:14

to Asia for many of our speakers

38:17

is costly and cannot be evenly borne

38:19

by all speakers. The conference needs to

38:21

have a cushion of at least a

38:24

hundred thousand US dollars in a bank

38:26

in order to ensure that people and

38:28

vendors are paid on time. and that

38:31

the conference is actually happening. The transition

38:33

organizing committee started discussing on how to

38:35

solve these situations. If you want to

38:38

get in touch with the program committee

38:40

or want to join that, like I

38:42

did, by the way, please contact committee

38:45

at Asia-Bisticon.org directly if you can help

38:47

by joining the Asia-Bistic program committee, recruitment

38:49

of speakers, or financing if you want

38:52

to sponsor. This is always the two

38:54

priorities when running a successful conference. So

38:56

the website will be updated this weekend

38:59

and they already have with these details

39:01

and the transition committee. We know this

39:03

may all come to a bit of

39:06

a shock to the community but we

39:08

believe that this is a welcome or

39:10

a good resent to the conference and

39:12

that makes it stronger in future years

39:15

and give it the ability to provide

39:17

excellent papers, talks and hallway discussions for

39:19

another 20 years. Yeah,

39:23

it's a bit of sad news

39:25

about the conference because I knew

39:27

a few people were warming up

39:30

to it and a lot of

39:32

people actually got in contact with

39:34

us to ask when we call

39:37

for papers to be able to

39:39

submit a paper to the conference.

39:41

So it's a bit of sad

39:44

news that's come about. There is

39:46

some people that have already purchased

39:49

tickets and I believe these conversations

39:51

about just having a little get

39:53

together in Tokyo for those that

39:56

are already committed to travel. So

39:58

I believe there was some, I

40:00

can't remember where I saw it,

40:03

I've had so much email on

40:05

the last week. Yeah, but I

40:08

already bought tickets. Yeah, I think,

40:10

I think. there is something happening.

40:12

We will look for that email

40:15

and bring it to you in

40:17

an upcoming release of the episode

40:19

if I can find it to

40:22

sort of read it out. So

40:24

if anybody out there is listening

40:27

and is already committed to going

40:29

to it, at least, you know,

40:31

we might be able to let

40:34

people know what some other people

40:36

are planning to do. So it's

40:38

a shame that this has happened

40:41

this year, but you know, we've

40:43

got to look forward. to 2026

40:46

now for Asia BSDcon and hopefully

40:48

we get that off the ground

40:50

and we can all get together

40:53

and catch up. I certainly would

40:55

like to get to 2026. Let

40:57

me tell you I miss going

41:00

to conferences and when I get

41:02

a chance to, especially the Asia

41:05

ones, they're close for me. So

41:07

yeah, I'm looking forward to getting

41:09

back to Asia in either in

41:12

Taiwan when they switch back and

41:14

forth again. Yeah, so fingers crossed

41:16

and good luck to the organizing

41:19

committee and let's get it on

41:21

for 2026. Yep. So that is

41:24

this news and we heard about

41:26

another conference interestingly, the ghost BSD

41:28

Desktop Conference. What's that about? This

41:31

is an article over on the

41:33

Fronix website Benedict and it says

41:35

the year of the BSD Desktop.

41:38

Question Mark. There's going to be

41:40

a BSD Desktop Conference at least.

41:43

And the story goes. Technical BSD

41:45

conferences aren't quite as common as

41:47

the many Linux conferences these days.

41:50

For the BSD conferences that do

41:52

happen, they tend to be more

41:54

general in nature than carrying a

41:57

Desktop Focus. But being announced this

41:59

week was Ghost BSDCon 2025. As

42:02

a forthcoming developer conference largely focused

42:04

on Desktop Use. of this free

42:06

BSD derived distribution. Ghost BSDCon 2025

42:09

was announced as a virtual conference

42:11

taking place 29 March 2025 on

42:13

the Berkeley Forest Free BSD Lodge.

42:16

While hosted by the Ghost BSD

42:18

project, the virtual conference will be

42:21

around developing, deploying and using desktop

42:23

systems based on any BSD variant.

42:25

This is the first time Ghost

42:28

BSDCon is being put on virtually

42:30

or otherwise. As a reminder, is

42:32

a free BSD derived operating system

42:35

employing the Marte Desktop for a

42:37

nice out-of-the-box BSD Desktop experience. Those

42:40

unfamiliar with ghost BSD can see

42:42

our many past ghost BSD articles

42:44

and visit ghost BSD.org. For those

42:47

wanting to learn more about ghost

42:49

BSDCon 2025 as the new BSD

42:51

Desktop Conference can find the initial

42:54

details via this week's call for

42:56

papers for this virtual conference. Okay,

43:00

yeah, so as a kind of

43:02

a in-between or replacement conference for

43:04

the age of obviously, this might

43:06

be for some people of interest

43:08

if they are in the desktop

43:10

area. Yeah, it sort of goes

43:12

hand in hand with this, currently

43:14

there's work going for a special

43:16

interest group for free BSD, so

43:19

you'll probably see on the free

43:21

BSD YouTube channel, there's been the

43:23

recordings for... the last two special

43:25

interest groups for desktop BSD. So

43:27

utilizing free BSD as a base

43:29

and building that base up to

43:31

be more desktop oriented for laptop

43:33

and desktop users. So there's a

43:35

fair bit of work that's happening

43:38

in the desktop space for free

43:40

BSD. It's sort of been a

43:42

bit of a side show in

43:44

the past because a lot of

43:46

developers either use Max or their...

43:48

The operating system is being used

43:50

for big deployments. you know, in

43:52

the big deployments that we've talked

43:54

about previously on the show, you

43:57

know, like Netflix's and the Playstations

43:59

and that sort of stuff. So

44:01

there's sort of, there hasn't been

44:03

this focus around the desktop, but

44:05

there's been a bit of a

44:07

renaissance with the direction that Linux

44:09

is going, people are looking for

44:11

alternatives to get. ownership of their

44:13

computer back. They, you know, every

44:15

time that, you know, all these

44:18

things, shims get put into different

44:20

operating systems. You look at the

44:22

Windows 11 show, it's just like,

44:24

you know, I was going to

44:26

use a swear word there, but

44:28

I'm trying to keep this G-rated.

44:30

The amount of stuff that they're

44:32

jammed in Linux, people want to

44:34

have a simply simple computer back

44:37

to just do computing things that

44:39

they used to do. And, you

44:41

know, free BSD. can be one

44:43

of those like lifeboats for people

44:45

to come back and reclaim control

44:47

of their computer to continue doing

44:49

doing what they need with their

44:51

computer. So it's there's some exciting

44:53

stuff happening and I'm looking forward

44:56

to seeing what their outcome of

44:58

that over the next 12 months

45:00

is going to be. So yeah

45:02

good on ghost BSD for putting

45:04

something on and getting getting that

45:06

you know that fire burning. Yeah

45:08

that's definitely good to have. And

45:10

here goes an article by our

45:12

very own Tom Jones recovering from

45:15

external Z-Rude on his own blog

45:17

adventurous dot me and Tom- What's

45:19

Tom- What is Tom doing something?

45:21

Oh geez, well done Tom. You

45:23

mean outside of B-S-D? He has

45:25

time. What is to be an

45:27

article about Dragonfly B-S-D-D? Yeah, that's

45:29

that's the next article. His installation

45:31

instructions for Dragonfly. Yeah, we're still

45:34

waiting for that. It's kind of

45:36

becoming a running gag on the

45:38

show. But let's jump to his

45:40

article now. So his favorite terrible

45:42

computer, a separate link to another

45:44

blog post he wrote, has been

45:46

locking up after the impact. and

45:48

I need to panic a VEM

45:50

a lot to find a firmware

45:53

crash. The hard power cycle reboot

45:55

loop is taking too long and

45:57

I'm done with it. More knowingly

45:59

is also hard locking up copying

46:01

the VEM off the machine. This

46:03

forced a trip to the attic,

46:05

a reinstall of my large memory

46:07

host which for reasons had a

46:09

128 gigabytes NVME which is silly

46:12

on a machine with 128 gigabytes

46:14

of RAM. That is fine. I

46:16

can pull the NVM and drop

46:18

it in an enclosure and drop

46:20

it in an enclosure. Just copy

46:22

locally. For a ZFS disk on

46:24

a machine with an existing Z-route,

46:26

this was turned out to be

46:28

very painful. Somehow no one has

46:31

written down how to do this

46:33

either. Normally Z-pull import takes a

46:35

name as an argument, but it

46:37

can take a pool idea as

46:39

well. If you're using a machine

46:41

with Z-FS en route, then you

46:43

will already have a pool called

46:45

Z-route. So trying to import your

46:47

pool from an external drive will

46:49

fail with the name already in

46:52

use. finding the pool ID can

46:54

be done with the D argument

46:56

to Z-pull import. It's a long

46:58

number if you have to type

47:00

it up. This takes a device

47:02

to use as a pool so

47:04

he has a test output from

47:06

a Z-pull import and that displays

47:08

the idea. The ID. Now you

47:11

can use the pool ID instead

47:13

of a name with Z-pull import.

47:15

So you can do Z-pull import

47:17

minus capital R slash M&T. So

47:19

this is the alternative root mount,

47:21

then capital N, then F. the

47:23

long ID of your pool and

47:25

then lowercase T old Z route.

47:27

With this command I ask the

47:30

C pool to import with a

47:32

different root location capital R so

47:34

I don't clubber my existing file

47:36

system do not mount anything minus

47:38

n or minus capital n and

47:40

use the ID and a temporary

47:42

name for the pool which is

47:44

the lowercase T old Z route.

47:46

With the pool imported you can

47:49

find your data set and mounted

47:51

with ZFS list and mount dash

47:53

T ZFS. Oh yeah good to

47:55

know about this. and in case

47:57

you needed it's available here in

47:59

our show notes for reference. Yeah, this

48:01

is this is a good guide. You

48:03

know, I haven't really thought too much

48:05

about to document this down. I have

48:07

I've had come across this particular problem

48:09

quite often when I've got to

48:11

move machines and things like that. This

48:14

makes it really simple for everybody to

48:16

follow along at home and hopefully will

48:18

help them out. It'll probably be

48:20

slerved up into an AI data set

48:23

before you know it. Just type

48:25

it into your generity of AI and

48:27

it'll tell you exactly what Tom

48:29

just said. Yeah, so if you see

48:31

this, this ID for a pool

48:34

come up somewhere in, as an example,

48:36

when ChatGBT gives you an

48:38

answer, and you know where

48:40

it's being ripped from. It's

48:42

Tub's pool, yeah. Yeah. We're

48:45

still laughing, but it's

48:47

all happening in the

48:49

background as we speak. Yes,

48:51

it is. So moving on

48:53

to the next article.

48:55

We've got a... It's a blog post

48:57

from JP Men's and it is

49:00

labelled Create a New Issue in

49:02

a GitHub repository with Ansible. So

49:04

Benedict and myself are big Ansible

49:06

fans because it achieves so much

49:09

and you don't have to remember

49:11

stuff. You can just run the

49:13

Ansible scripts again and fixes your

49:16

problems. So it goes. I

49:18

want to create a new

49:20

issue on a GitHub repository

49:22

using Ansible. GitHub Rest API

49:24

Docs have a sample which

49:26

works with Curl which I

49:28

adapted for Ansibles, URI

49:30

module. And there's some

49:32

playbook code there for Ansible.

49:34

And I just like hard

49:36

coding secrets in codes so

49:39

I use a lookup which

49:41

tends to read the token

49:43

I'm using. So I read

49:45

the token as the actual

49:47

playbook plays. and the result

49:49

is expected. So then he's

49:51

got the issue opened and

49:53

the example of what it looks

49:56

like on the web page for

49:58

GitHub. So note labels. is an

50:00

array of labels and new labels

50:02

are automatically added by the API.

50:04

And that's the end of the

50:06

article, but it's really good. It's

50:09

good to see that, you know,

50:11

you can just like bang some

50:13

stuff out, query quickly, into other

50:15

platforms like GitHub, to, you know,

50:17

from directly from the command line

50:19

of your free BSD or OpenBSD

50:21

or whatever BSD machine. So, well

50:23

done. Okay, then we have... Ted

50:25

Unanks, which we haven't covered in

50:27

a long while. I haven't heard

50:30

from Ted ages. He disappeared off

50:32

Master Don with his honk creation

50:34

and he's back. He's going to

50:36

post. His block flag, Ted Unings,

50:38

has been, you know, active, but

50:40

I think we did such a

50:42

good job of covering some of

50:44

his latest exploits. And so this

50:46

one may be a reintroduction to

50:48

Ted. And so this goes to

50:50

stories I refuse to believe. The

50:53

internet is filled with stories that

50:55

purport to teach us valuable lessons

50:57

or something about how the world

50:59

works, and they're really important because

51:01

they really happened. NASA spent millions

51:03

of dollars designing a space pen,

51:05

which was really foolish when they

51:07

could just use the pencil like

51:09

the Russians. I think not as

51:11

many people believe that anymore, but

51:13

it's still floating around out there.

51:16

Here are three more storage which

51:18

I will never decompose, because it's

51:20

more fun to retell this tale

51:22

than question why. The first is

51:24

third burgers. In the mid-80s, restaurant

51:26

chain A&W launched a third-pound burger,

51:28

sorry a third-pound burger, to compete

51:30

with McDonald's. But the initiative failed

51:32

because Americans are too stupid to

51:34

understand fractions and thought four is

51:36

bigger than three. Oh, okay. I

51:39

think people like the story because

51:41

they knew fractions or they know

51:43

fractions. So if everyone else gets

51:45

it wrong, that affirms they're smarter

51:47

than everybody else. the TED words.

51:49

So the source for the story

51:51

is a paragraph or two in

51:53

top of it. 2007 autobiography written

51:55

more than 20 years after the

51:57

events in question. He was the

51:59

owner at the time and inventor

52:02

of the shopping mall. He was

52:04

also convicted of price fixing in

52:06

a separate incident, which is one

52:08

part at Ominam and one part

52:10

billionaire which I will say or

52:12

do anything to be right. It's

52:14

not the product which sucks. It's

52:16

the customers who is stupid. For

52:18

context... A&W had been not just

52:20

declining, but dying when top men

52:23

took over the business. By the

52:25

time customers allegedly had to do

52:27

fraction math, most of them probably

52:29

had their own opinions. I also

52:31

questioned a business acumen of only

52:33

doing focus group testing after rolling

52:35

out this all-in-sale strategy. Wouldn't you

52:37

first do some local testing before

52:39

going nationwide? The biggest question I

52:41

have is, given your recent discovery

52:43

of Americans in numeracy, how do

52:46

you not pivot to selling fifth

52:48

pound burgers. And I can totally

52:50

believe they did a focus group

52:52

and one dump said that burgers

52:54

sounded too small. It's a funny

52:56

anecdote and might have bubbled up

52:58

to the CEO. But what did

53:00

the other 99 people in the

53:02

survey say? Then the next story

53:04

is about Target Dad. Target ran

53:06

an ad program to so detailed

53:09

they would know you're pregnant before

53:11

you're pregnant before you're pregnant. An

53:13

outrage and outrage Dad storms into

53:15

the store manager's office demanding to

53:17

know why they're trying to hypnotize

53:19

his high school daughter into getting

53:21

pregnant with diapur ads. Then a

53:23

few days later he apologized. Target

53:25

was right. She was pregnant. What

53:27

sets me off is the Disney

53:29

perfection of the story. Dude comes

53:32

into the restore to complain happens

53:34

every day. But the story would

53:36

be meaningless without the follow-up apology

53:38

to confirm the truth and then

53:40

the mouse pulled the thorn from

53:42

the lion's paw. Oh wow, that's

53:44

a nice reference there. Peepback at

53:46

Bistenado TV if you know what

53:48

the reference is about. The story

53:50

source wasn't the manager, however. It

53:53

was the assistant to the regional

53:55

manager or something. something who somehow

53:57

happened to be hanging out in

53:59

the manager's office for both events.

54:01

Don't they have their own job

54:03

to do? And recorded the conversations

54:05

to provide verbatim quotes to the

54:07

reporter. I wouldn't expect the flyer

54:09

from Target to have been preserved,

54:11

but the article is pretty vague

54:13

on the dates when the super

54:16

specific ad program was running or

54:18

when the dad got mad. Did

54:20

the assistant? to the regional manager

54:22

affect the program change? It's supposed

54:24

to be a cautionary tale but

54:26

who canceled the program before the

54:28

tale was told. It seems more

54:30

like one more hypothetical in an

54:32

article filled with hypotheticals that got

54:34

lost in the telling. The whole

54:36

article is built up on tales

54:39

of the dark secrets of advertising,

54:41

but when it comes to verifying

54:43

there is no confirmation or anything.

54:45

Even the author's original source has

54:47

gone silent and after feeding him

54:49

a bunch of stories. There's also

54:51

what I'll call the null hypothesis

54:53

story. This incident could have happened

54:55

even without any tracking. Mail coupons

54:57

entirely at random, you're going to

54:59

mail some diaper coupons to teenage

55:02

girls. We should therefore expect some

55:04

mad dads regardless, but this one

55:06

anecdote doesn't tell us if we've

55:08

exceeded the mad date base rate.

55:10

So that's another reason I think

55:12

it's made up. If your research

55:14

complaints received by target managers, you'd

55:16

unearthed so much in saying crap

55:18

this story would never register. Then

55:21

there is the brown candy story. Very

55:23

small detail in your contract, then check

55:25

if the vendor did it and you'll

55:28

know if they did the rest. People

55:30

really, really love this story. But did

55:32

it happen? Well, of course, look, here's

55:34

a photocopy of the contract. But did

55:36

it work? I'm supposed to believe that

55:39

upon discovering a brown candy, when Halen

55:41

went out and double checked all the

55:43

pyrotechnics and whatnot. When? How much time

55:45

is there between arriving at the backstage,

55:47

ready room and their performance? Isn't the

55:50

opening act already out there? How long

55:52

did they delay the show? If the

55:54

check was that fast, why not just

55:56

always perform it? Why would anyone... assume

55:58

that the snack subcontractor was or has

56:00

any relation to the munitions subcontractor. This

56:03

is like the age-old advice that you

56:05

should never buy a lawnmower from salesman

56:07

with a crook tie. There's a bit

56:09

more to this story, but the overall

56:11

at the end summary of Ted is

56:14

recency. What all these stories have in

56:16

common is they came to like to

56:18

explain something that happened long ago. By

56:20

the time the story was out the

56:22

sources were hard to find and who

56:25

really cares. But did you hear about

56:27

the Air Force AI drone that turned

56:29

around and killed its operator which turned

56:31

into a simulation and then it was

56:33

a thought experiment that it was just

56:36

a made-up story? Or the botanet of

56:38

Bluetooth toothbrushes that destroyed a bank? That

56:40

was pretty quickly clarifying to a hypothetical

56:42

of a mistranslation of a miscommunication. What's

56:44

different is that these events allegedly occurred

56:47

now and people asked questions. If either

56:49

of those examples appeared in a book

56:51

published 20 years from now, they'd be

56:53

accepted without question. Oh yeah, cybersecurity was

56:55

wild back in the 20s. I remember

56:57

that Toothbrush Botnet took down a lot

57:00

of our services. Instead, we'll be reading

57:02

a book about refrigerator botnets in 20

57:04

years and somebody will question if it

57:06

really happened and the internet will tell

57:08

them of course it happened. That's where

57:11

Silicon Valley got the inspiration got the

57:13

inspiration from. Okay.

57:15

Honorable mention, low background steel. Yeah.

57:17

Yeah, good to see. Ted doing

57:20

his reading that might not be

57:22

computers, but it's good to see

57:24

him back blogging. Yeah, he also

57:26

brought a couple of articles about

57:28

coding and a new grab implementation

57:31

that doesn't suck, but that is

57:33

up to you to discover on

57:35

his blog role. Yeah, when you

57:37

started this post off, I couldn't

57:40

stop. stop laughing. And I just

57:42

couldn't, I was having a private

57:44

chuckle themselves. It was quite funny.

57:46

The third burger set was gold.

57:49

But yeah, interested, it brought back

57:51

memories about NASA spending millions of

57:53

dollars of trying to make a

57:55

space pen when they could just

57:58

follow the Russian cheese pencil. Well,

58:00

many decades ago I did scuba

58:02

diving. And you know, when you

58:04

take your tablet down, not your

58:07

tablet, your iPad tablet, no, no,

58:09

no, just your writing tablet. All

58:11

we used was just the graphite

58:13

to be pencil and you could

58:15

write messages to each other and

58:18

it was fine. It was like,

58:20

there was nothing special about it.

58:22

So it's like, you know, pencil does

58:24

the job. Why invent something

58:27

else? Different environments. Yeah.

58:29

News Roundup of the week

58:31

is a commit message. So

58:33

this is around the January

58:36

19, 2038 date limit in

58:38

UFS-1 file systems to February

58:40

7, 2021-6. So we can

58:43

guess who committed this to

58:45

the source tree and it

58:47

was Dr. Kirk McCusick and

58:49

it happened on the 31st

58:52

of January 2025. So it

58:54

was some code committed to

58:56

get over the 2038. issue

58:59

that was still existing in

59:01

the UFS1 code and it's

59:03

to bump time T from 32

59:05

bit to 64 bit so at

59:07

least you'll be able to use

59:10

UFS1 now and not have

59:12

to worry about it now

59:14

lifetimes so it'll be somebody

59:16

else's problem so well done

59:18

Kirk and it's good to

59:20

see that they still love

59:22

being given to the older

59:25

file systems that exist. with

59:27

inside our favorite BSDs. Yeah,

59:29

so UFS-1, got this fixed, UFS-2

59:31

is safe, is fine with that

59:33

date problem, and it's good to,

59:36

again, have the little bit of

59:38

love given to that old version,

59:40

if it's still around somewhere and

59:42

people cannot upgrade for whatever reason

59:45

to UFS-2, so they also get

59:47

this date extension. BSD now was

59:49

sponsored by Tar Snap. Everyone

59:51

needs backups, and Tarzap ensures

59:54

that your backups are not

59:56

only safe. but also secure. Your data

59:58

is encrypted on your device before. for

1:00:00

being sent to the cloud, you can

1:00:02

be sure that only you have the

1:00:04

ability to read your data. Tarsnap takes

1:00:06

your data and works out what is

1:00:08

duplicated. It then assembles the data into

1:00:10

compressed blocks and cricks them with your

1:00:12

local private key, and this key never

1:00:14

leaves your system. The data is then

1:00:16

uploaded into the cloud. Even if someone

1:00:18

is able to obtain your data in

1:00:20

the cloud, they will not be able

1:00:22

to decrypt it and access your files.

1:00:25

Tarsnap is easy to use. Tarsnap is

1:00:27

prepaid so you never have to worry

1:00:29

about an unexpected bill. Tarsnap is fully

1:00:31

open source allowing you to inspect the

1:00:33

code and make sure it does what

1:00:35

we say it does. Tarsnap has bug

1:00:37

guarantees so that if you find errors

1:00:39

in the code you can get paid

1:00:41

for helping make the software better. With

1:00:43

clients and all major platforms there's no

1:00:45

excuse not to have good backups. Go

1:00:47

to Tarsnap.com to learn more. And

1:00:56

that brings us to the end

1:00:58

of this episode. We have, oh,

1:01:00

ha. Almost to go. We have

1:01:03

feedback. Yes, feedback. I slipped this

1:01:05

into the show notes. We're all

1:01:07

getting ready for the show and

1:01:10

I said the Benedict, where's the

1:01:12

feedback? There he goes, what feedback?

1:01:15

And I go, well, we got

1:01:17

email from Nelson. It's like, where's

1:01:19

the feedback? And he's gone. I

1:01:22

haven't included. So, you know, madly.

1:01:24

madly with my non-get skills trying

1:01:26

to put it into get hub

1:01:29

for the show notes but we

1:01:31

got it into the show notes

1:01:33

so Nelson BB has provided some

1:01:36

feedback and it goes like this

1:01:38

comments in the latest episode of

1:01:40

BSD now indicate puzzlement over the

1:01:43

status of GCC support for utter.

1:01:45

The issue is that two decades

1:01:48

ago about GCC four or so

1:01:50

parts of the GCC compiler for

1:01:52

Ada were written in Ada itself.

1:01:55

This meant that you need an

1:01:57

earlier version of GCC to build

1:01:59

the newer compiler and the bootstrap.

1:02:02

problem exists ever after. This has

1:02:04

also happened more recently with the

1:02:06

sport for the D language. If

1:02:09

OS distribution always supplied all languages

1:02:11

that GCC supports this would not

1:02:13

be an issue. However, some major

1:02:16

distributions failed to do that. Among

1:02:18

them, CentoES, Gunew Linux, and some

1:02:21

members of the BSD family. Free

1:02:23

BSD has a package GCC6 ox.

1:02:25

2.1080516 underscore 3.1-comer that provides ADAR

1:02:28

but is a nine major compiler

1:02:30

generation behind, making it a difficult

1:02:32

job to bootstrap ADA support for

1:02:35

modern compiler versions. Modular 2, GM2

1:02:37

has been supported by GCC for

1:02:39

a few years now, but many

1:02:42

distributions fail to supply it. At

1:02:44

Utah, I routinely build new GCC

1:02:46

releases every few weeks on our

1:02:49

major servers and my goal is

1:02:51

always to include every compiler that

1:02:54

GCC can produce. Apple does not

1:02:56

supply any GCC compilers and it

1:02:58

has been several years since out-of-the-box

1:03:01

GCC releases will build on MacOS.

1:03:03

There is some work in this

1:03:05

area to fix that problem but

1:03:08

it has not been integrated into

1:03:10

the master GCC sorcery. Yeah, interesting.

1:03:12

It's good to see that people

1:03:15

still dive into the weeds with

1:03:17

compilers to keep them going because

1:03:19

you know if it's clear here

1:03:22

that you know if you've got

1:03:24

a gap you're not going to

1:03:27

be able to bootstrap the newer

1:03:29

ones because you need the older

1:03:31

stuff the bootstrap the newer stuff

1:03:34

so thanks very much for that

1:03:36

Nelson and the highlighting the Ada

1:03:38

stuff in itself. This, I think

1:03:41

I've mentioned on previous episodes, this

1:03:43

podcast called Advent of computing and

1:03:45

the old compilers. and Ada have

1:03:48

been mentioned on the podcast in

1:03:50

the past. So if you're bored

1:03:52

with listening to us, which I

1:03:55

hope you're not, and after this

1:03:57

podcast, you can go and check

1:04:00

out Adventive computing and give them

1:04:02

a sub and have a look

1:04:04

at the back catalog. Yep, there's

1:04:07

some good stuff in there and

1:04:09

good training ground if you want

1:04:11

to flex your programming muscles. And

1:04:14

it's always good to say, look,

1:04:16

I solved a couple of those,

1:04:18

gave me a bit of thinking

1:04:21

time. and I have a much

1:04:23

better understanding of a computer science

1:04:25

problem if you want to frame

1:04:28

it that and overall you have

1:04:30

a better approach to these kinds

1:04:32

of problems or at least see

1:04:35

what issues are out there that

1:04:37

people can solve and that is

1:04:40

really now the end of this

1:04:42

episode really really that is that

1:04:44

is that now it is it

1:04:47

is but before we before we

1:04:49

go don't forget to check our

1:04:51

show notes out at bsd now

1:04:54

dot TV and if you've got

1:04:56

any questions or feedback or suggestions

1:04:58

for the show please send them

1:05:01

into feedback at bsd now dot

1:05:03

TV and we will either get

1:05:05

back to you very quickly on

1:05:08

the email or get them prepared

1:05:10

ready for the show so we

1:05:13

can read them out on air.

1:05:15

Yep enjoy the week until next

1:05:17

time we have an episode spoiler

1:05:20

alert next week and There we

1:05:22

go, enjoy. Catch you later.