<srhb>
I'm hoping their AMD models start getting way better in general...
<etu>
i3 that supports ECC
<etu>
qyliss^work: But they are very rare
Guanin has joined #nixos-chat
<emily>
(#nixos ->) you can dodge irccloud bans if you NAT the connection through your server :V
* emily
just got that set up yesterday
<gchristensen>
O.o
<gchristensen>
do user cloaks also work?
<emily>
yes, though I don't think you need a proxy for that
<emily>
it also supports acting as a bouncer and connecting with another IRC client with backlog replay, btw
<emily>
so you can use it just for the mobile IRC support if you want
* etu
just uses weechat and the weechant relay protocol
<emily>
doesn't get you mobile notifications, alas
<gchristensen>
weechat user scripts and pushover? :P
<emily>
yeahhhh...
<gchristensen>
some things I just don't want to do
<emily>
it'd only be 5x as much work for like a 2x jankier user experience
<emily>
I'm okay paying irccloud, they're cool, they use erlang :p
<emily>
(how Discord manages to be such an unreliable pile of crap while being written in Erlang is a question that confounds me on a daily basis)
<gchristensen>
LOL
<simpson>
emily: In a word, it's the culture.
<simpson>
(Two words: *Gaming* culture.)
<gchristensen>
I used to pay for but not use irccloud to support a co based around IRC...
<emily>
I kind of wanted to switch everything over to Matrix and use that for IRC too instead. still not totally sure I won't do that, but so far it's seemed like substantially more fuss and wonkiness than IRCCloud + Signal for anything I actually care about being e2e...
evanm has joined #nixos-chat
<etu>
emily: yeah it does
<etu>
emily: But not very good :D
<emily>
fair enough :p
<emily>
Quassel is another option I've considered. not looked into it very much.
<gchristensen>
hoo buddy travisci just laid off a mess of people
<__monty__>
Dang, any news on what happened?
jasongrossman has quit [Quit: ERC (IRC client for Emacs 26.1)]
<qyliss^work>
Saw that coming when they got acquired a month or so ago :(
<joepie91>
averell: got famous on the back of their 'open-source' ExtJS thing, then gradually started closing pieces off to non-paying customers
<joepie91>
I believe they're now at the point where /technically/ you can still use the ExtJS bits under a GPL license, but you have to enter all your personal data to get a download, and there's no public repo, just patches posted on a forum
<joepie91>
and none of the other bits (with ExtJS being designed to *need* those bits to be ergonomically useful) are open-source of course
<averell>
starting to be a trend, with mongo, emby etc
<joepie91>
aye
<joepie91>
turns out, "let's make open-source stuff with VC money and we'll figure out how to make money later!" isn't a great business model... whodathunk
<joepie91>
<.<
<joepie91>
(thus also my general dislike of commercial open-source, very few companies pull it off in a way that doesn't disadvantage the public commons)
jasongrossman has quit [Ping timeout: 255 seconds]
<__monty__>
joepie91: Red Hat being acquired by IBM means there's very few if any succesful open source business model examples though. Who'd you say "pulls it off?"
endformationage has joined #nixos-chat
<joepie91>
__monty__: Sentry seems to be doing quite well, for example
<__monty__>
joepie91: What would you say they do right/different? At first sight it looks like maybe they're just tackling a hairy problem no one wants to deal with, reducing competition.
<joepie91>
__monty__: their code is unconditionally open-source, yet they have what seems like a viable business
<joepie91>
the exact reasons for that, I can't list exhaustively :)
<joepie91>
because I'm not privy to their internal business details
<kgz>
gchristensen: rip
<kgz>
they had a good run
<joepie91>
__monty__: sorry, was otherwise distracted for a bit. to continue on the above... they're not trying to sell open-source software, they sell a service where the backing system happens to be open-source
<joepie91>
the sales of the paid service presumably back the support of the open-source system
<joepie91>
financially
<joepie91>
from the perspective of the business, the software is totally irrelevant for what they're selling; namely, a turnkey error tracking and analysis service... the software is just an implementation detail
<joepie91>
and 'turnkey' is important there, because that would be a major reason why people would pay for the service instead of just deploying the software themselves; it's one less moving part to deal with
ninjin_ has quit [Remote host closed the connection]
ninjin_ has joined #nixos-chat
__monty__ has quit [Quit: leaving]
averell- has joined #nixos-chat
averell has quit [Ping timeout: 252 seconds]
averell- is now known as averell
averell has quit [Quit: .]
disasm has joined #nixos-chat
__Sander__ has quit [Quit: Konversation terminated!]
<elvishjerricco>
Is samba just super slow for transferring lots of small-ish files (about 3M each)? These two machines are connected by gigabit ethernet, but it's only transferring at like 10 MB/s
averell has joined #nixos-chat
<joepie91>
elvishjerricco: yeah, that's a common issue afaik
<joepie91>
I've heard a lot of people complain about that before
<joepie91>
if at all possible, prefer rsync for large amounts of small files
<joepie91>
it'll likely hit line speed or, in the case of HDDs, HDD speed
<elvishjerricco>
Hm. Well the application is macOS Time Machine on network storage. It works by creating a sparse disk image file and formatting it to HFS+ (because Time Machine can't use any other FS). Dunno if Time Machine supports anything better than Samba.
<cransom>
i use afpd/netatalk for my time machine backups over the network and it's zesty enough
<gchristensen>
same, elvishjerricco, have you seen my post on it?
<elvishjerricco>
Seems to be working. Though not substantially faster unfortunately :/
<gchristensen>
so gigabit, 1,000 megabit-per-second should be doing 125 Megabytes-per-second, and you're at 10megabytes-per-second?
<gchristensen>
maybe check the network interfaces of the two and ensure they both agree they're at 1,000? /sys/class/net/INTERFACE/speed
<gchristensen>
elvishjerricco:
<elvishjerricco>
gchristensen: Well this is getting weird... I'm using a tinc network, and /sys/class/net/tinc.DAEMON/speed says 10... not so good... But when I test the speed of the tinc connection with iperf, I get 200 Mbits/sec. But when I test over the LAN, skipping the tinc, I get 928 Mbits/sec
<gchristensen>
sounds like we solved it then
<gchristensen>
tinc is lying about it speed
<gchristensen>
and incurs signifcant overhead... is it properly going point-to-point or is it leaving your network, would be my next question
<elvishjerricco>
gchristensen: It is definitely not leaving the network. My internet is not nearly fast enough for 200 Mbits/sec :P Regardless, why would I be getting like 10MB/s for time machine, but (200 / 8) MB/s for iperf between the same two machines?
<gchristensen>
is timemachine establishing a single connection, or just one?
<gchristensen>
is timemachine establishing a single connection, or many?
<gchristensen>
another thing to look at is look at `top`'s `wa` output, an `iotop`
<elvishjerricco>
Not sure...
<gchristensen>
just to make sure you're not disc-bound
<elvishjerricco>
gchristensen: Definitely not bottlenecked by the disk.
<gchristensen>
when writing a backup, what does top reveal? any process pegged at 100%? memory pressure? iowait? what does `pv /dev/urandom > /path/to/where/time/machine/writes/test` reveal?
<elvishjerricco>
gchristensen: So when I ran `dd if=/dev/zero | ssh host dd of=/dev/null status=progress`, I got 200Mbit/sec, and tincd was taking 45% of a CPU core.
<gchristensen>
that is a decent test
<gchristensen>
try urandom
<elvishjerricco>
The pv thing was only going at 24MiB/s, but I think that was bottlenecked by urandom, since outputting to /dev/null instead didn't make it any faster
<elvishjerricco>
(local /dev/null)
<gchristensen>
my concern is tinc implements compression
<gchristensen>
so zeros might go real fast when random data might no'
<elvishjerricco>
gchristensen: Hm, and the data being written by Time Machine is encrypted
<gchristensen>
so if you have a big already compressed file like an ISO or something
<joepie91>
elvishjerricco: write random to a file, then pipe the entire file through at once?
<joepie91>
to get around the urandom bottleneck
<gchristensen>
+1
<elvishjerricco>
joepie91: Yep, already started that :)
<joepie91>
ah, ok :)
<elvishjerricco>
When reading from urandom, will the kernel or the process be blamed for the CPU usage? Looks like the process but that seems odd to me
Guanin has quit [Ping timeout: 258 seconds]
<elvishjerricco>
Ok copying a 4G randomized file to the netatalk share directly was going at 88MiB/s. My dd over ssh thing only goes at 38MB/s
<elvishjerricco>
Oh, I guess I don't know that the netatalk share is being reached via tinc... It's probably using the local network.
<elvishjerricco>
Which makes this "18 hours remaining" make even less sense...
<infinisil>
I'm wondering, man pages have existed for so long already, but they're really basic
<infinisil>
Is there something new that is like man, but has more features?
<manveru>
info :)
<infinisil>
Oh!
<manveru>
not that anyone is using it...
<infinisil>
"Unknown command (k)"
<elvishjerricco>
Hm. The Mac's CPU is pretty much pegged at a load average of 4-5 (the machine has 4 cores / 8 threads). I wonder if that's part of the problem
<joepie91>
infinisil: Firefox!
<joepie91>
(only half kidding)
<infinisil>
It might be usable with vim movement
<manveru>
i use it via spacemacs
<manveru>
so i get evil binds
<elvishjerricco>
The CPU flares up, and a bunch of data gets transmitted, then it goes back down, and the transfer stops going up
<infinisil>
joepie91: Can't use firefox on a headless server, it should really be something in the terminal
<joepie91>
infinisil: lynx :P
<infinisil>
(well you technically can, but we'll ignore X11 forwarding..)
<infinisil>
Eh..
<infinisil>
But yeah info is kinda cool
<infinisil>
I'll try see if there's vim movements for it
<joepie91>
I mean, isn't this essentially what NixOS does?
<joepie91>
for its manual on the live CD
<infinisil>
Oh I've never opened that
<infinisil>
Does it start the browser?
<joepie91>
it uses some sort of terminal-based browser
<joepie91>
afaik
<manveru>
i thought it uses w3m
<joepie91>
I don't mean the shortcut
<joepie91>
I mean the ctrl+alt+f6 or w/e
<joepie91>
manveru: quite possibly
<elvishjerricco>
And mdworker_shared is taking up a lot of CPU along side it, despite me adding the share to spotlight's ignore list
<infinisil>
Goddamnit but how am I supposed to find info on "info" online..
<infinisil>
That's like the worst possible name in that sense
<manveru>
infinisil: --vi-keys
<joepie91>
ha
<manveru>
try that :)
<infinisil>
Oh nice
<manveru>
searching "gnu info" usually helps
<infinisil>
And now with --vi-keys I can't open pages with enter
<manveru>
no?
<infinisil>
Enter just scrolls the page by one line for me
<infinisil>
manveru: Does it open the links for you?
<manveru>
yeah
<infinisil>
Damnit
<infinisil>
Also, small "h" doesn't "start the info tutorial" in vi mode, it shows all info commands, which should be "H"
<infinisil>
I guess I won't be using info lol
<manveru>
yeah, i never use it in the terminal, so no idea man :|
<infinisil>
There apparently is an emacs viewer too, maybe I'll try that
<manveru>
as i said, with spacemacs it's fine :D
<infinisil>
Ah sorry I missed that
<infinisil>
Yeah I have my emacs configured with evil mode too, seems to work just fine :D
evanm has quit [Ping timeout: 256 seconds]
lejonet has quit [Ping timeout: 246 seconds]
lejonet has joined #nixos-chat
* colemickens
really likes the looks of 'niv'
* colemickens
might move three or four things to it
<emily>
niv?
<colemickens>
I leaned about it on discourse, but looked into it after it was featured in NixOS Weekly
<colemickens>
a better example is where I use it to maintain an overlay of dozen(s?) of packages that are updated from git sources sometimes multiple times a day: https://github.com/colemickens/nixpkgs-wayland
<colemickens>
I'm interested in using `niv` and dropping my own update.sh scripts, which are already diverging in the few places I have used this sort of setup
<manveru>
ah, ok
<manveru>
i don't really have anything that complex :)
<joepie91>
hm. what is "a Nix project" in the context of that readme?
<manveru>
i tried niv a few weeks ago when it was first published, but couldn't find any use for it, so i kinda forgot about it
<elvishjerricco>
Think I found where my bottleneck lies... Although I can write to the share with high speed, writing to the HFS formatted sparse image file that time machine uses is very very slow.
drakonis has joined #nixos-chat
<elvishjerricco>
After a bit of testing, yea these sparse image bundles are about a quarter the speed to write to. Guess that explains it :/
<elvishjerricco>
throttle each other fighting for bandwidth.
<elvishjerricco>
It doesn't really make sense though. As far as I can tell, writing to a sparse bundle is about 25% as fast as writing to the underlying FS (I've tested on the netatalk share and on the local disk), which implies it's not a CPU bottleneck. But the actual network usage is also 25% of what it is when I write to the share directly. And when I write to multiple sparse bundles on the share at once, they do seem to
<etu>
joepie91: Where do I need to drive to get there?
<joepie91>
etu: The Hague :D
<joepie91>
(fill rate and flavor distribution varies depending on how long it's been since the last order :P)
<etu>
joepie91: My main supplier of Club Mate in Stockholm stopped selling it last year :(
<joepie91>
etu: well okay, technically Leidschendam, but that is practically indistinguishable from The Hague
<etu>
Clearly I bought too little
<joepie91>
heh
<etu>
But I think it needs to be a team effort to make store keep products in stock :p
<joepie91>
we buy enough at a time that the supplier in Germany will drive a truck over to us to deliver it
<etu>
:D
<joepie91>
we also generally have winter mate year-round
<joepie91>
we just buy an absolutely giant batch when it's in stock again, just enough to last until the next availability cycle
<joepie91>
downside: those crates need to go somewhere!
<emily>
hmhm, i should visit
<joepie91>
etu: but yeah, the space on that picture was previously used by another tenant of the building, but they've left, and now the hackerspace is renting it and subletting it to people who want to rent cheap storage space... so our mate crates now double as a partitioning wall between the entrance to the hackerspace, and the storage area :P
<etu>
It's not a day-trip down there though :/
<joepie91>
emily: do!
<joepie91>
we're open to visitors whenever there's people around, which in practice means daily :P
<joepie91>
next 'open day' with events and stuff will be on march 30
<joepie91>
etu: you could always take a holiday to NL and do a hackerspace tour :P
<joepie91>
there's quite a few within easy travel distance of each other
<joepie91>
one hackerspace recently organized a 'road trip' to ~all hackerspaces in NL, visiting them in two days
<emily>
*nods* hopefully the conjunction of near the hague /\ feeling unanxious enough to socialise will be true at some point soon!
<joepie91>
emily: if it helps any, we do have a lounge (with a closable door and default-soft lighting and couches) that you can retract into when you need some time alone :)
<gchristensen>
on one hand buildRustPackage is slow as molasses to iterate on. on the other hand, carnix is hard to make work :|
<elvishjerricco>
gchristensen: I tried carnix like one time and gave up because it didn't handle git deps right :P
jackdk has joined #nixos-chat
<andi->
whats the current roadblock? I started hacking on it and will likely continu. If things ever get merged..