freenode/#devuan/ Wednesday, 2019-06-19

eyalrozHello all. Running beowulf (with user dir created in ascii), evince is giving me a lot of the following errors:16:58
eyalrozunable to create file '/home/my_user_name_here/.cache/dconf/user': Permission denied.  dconf will not work properly.16:58
eyalrozBut the permissions both for the file and the directory seem to be fine (a bit on the stingy side, i.e. owner-only). Why am I getting this?16:58
gnarfacenot sure but it could be an issue with the old config for the program17:00
gnarfaceeyalroz:17:01
drawkuladirectory dconf is 700 here and the file inside is 64417:01
gnarfaceoh sorry i didn't read carefully enough, it actually says permission denied...17:01
gnarfaceweird17:01
drawkulaminimal ascii install, upgraded to beowulf, then added xfce and more17:01
drawkulaso dconf came from beowulf, not via ascii17:02
eyalrozdrawkula: For me, it's 700 and 60017:02
xinomilofile here is 600 , dir 755, same install w/ mate17:02
eyalrozOh, btw, I use Cinnamon17:03
fsmithredI have 755 on dir and 644 on file17:03
drawkulawhat a diversity...17:04
fsmithredxfce here, and no evince17:04
eyalrozby the way... it's not that evince doesn't work17:05
eyalrozit just fills the console with dconf errors17:05
drawkulalots of chars to recycle... \o/17:06
eyalrozI also get:17:06
eyalrozAttempting to store changes into '/home/my_user_name/.local/share/recently-used.xbel', but failed: Failed to create file “/home/my_user_name/.local/share/recently-used.xbel.2SYC3Z17:06
eyalroz”: Permission denied17:07
eyalrozAnd /home/eyalroz/.local/share/ is 75517:07
eyalrozDo I actually need those folders to be world-writable? That doesn't seem right.17:07
fsmithredHow the hell do you save a file in evince these days???17:08
fsmithredMy .local/share is 70017:09
fsmithredand so are about half the directories inside it17:09
drawkulatouch ~/xyzzy17:11
drawkulais your home readonly?17:11
fsmithredok, I figured out how to save a file. It works here and dconf/user got updated.17:12
eyalrozdrawkula: Are you asking me?17:13
drawkulayes17:13
eyalrozMy home dir is not read only, and xyzzy has permissions 064417:13
drawkulaok17:13
drawkulathen I'm out of ideas17:14
eyalrozis it possible that evince or some dconf-related process changes its uid ?17:14
fsmithredcreate a new user and test. That will narrow down the problem.17:21
eyalrozfsmithred: Somehow  useradd is not populating my new user's home directory...17:26
fsmithredadduser phred17:30
drawkulauseradd ... -m ...17:30
drawkulabut manually copying /etc/skel contents should do too17:31
eyalrozdrawkula: Yes, thanks17:31
eyalrozfsmithred: http://paste.debian.net/1088497/17:31
eyalrozI ran evince from a terminal this time17:31
eyalroz(although was using an X session opened for my original user)17:32
eyalrozI meant, ran it from one of the Alt+Fn virtual terminals17:32
eyalrozIt seems there's some kind of apparmor policy in play here.17:33
fsmithredeyalroz, did you know you're using apparmor?17:39
fsmithredmy point is: do you need it?17:39
eyalrozfsmithred: Nope, had no idea.17:39
eyalrozDon't remember ever enabling it17:40
fsmithreddpkg -l |grep apparmor17:40
fsmithredaptitude why apparmor17:40
fsmithredall I have is one library17:40
eyalrozfsmithred: First I'll apt-get install aptitutde ...17:40
eyalrozi   linux-image-amd64          Depends    linux-image-4.19.0-5-amd6417:41
eyalrozi A linux-image-4.19.0-5-amd64 Recommends apparmor17:41
eyalrozThe kernel wants this? :-O17:41
fsmithredoh, I turn off Recommends17:41
eyalrozfsmithred: But it's on by default, right?17:42
fsmithredyeah17:42
eyalrozWell, let's see what happens when I remove apparmor17:42
fsmithredapt-get --no-install-recommends install blah17:42
fsmithredor set it in /etc/apt.conf.d/17:42
fsmithred that's /etc/apt/apt.conf.d/17:43
eyalrozI like recommendations...17:43
eyalrozexcept for this one17:43
fsmithredlol17:43
eyalrozBut apparmor - is it a daemon? a kernel module?17:43
fsmithredit's nice to review them before they get installed. You might want some but not others.17:43
eyalrozI'll go along with the herd17:43
fsmithredif it's a daemon, there should be /etc/init.d/apparmor or something like that17:44
fsmithredif you're going to keep it, you'll need to adjust it so it will let you do what you want.17:44
fsmithredI've never used it and know nothing about it.17:44
eyalrozNo, I'm removing it17:44
eyalrozbut17:44
eyalrozit doesn't quite die17:45
eyalrozI still get apparmor messages after having removed it17:45
fsmithredps ax |grep apparmor17:45
eyalroznothing.17:46
fsmithredmaybe you need to log out and in?17:46
eyalrozfsmithred: Restarting the x server maybe.17:47
* eyalroz will leave for a short while...17:47
fsmithredor even 'init 1'17:47
eyalrozfuck it, I'll just reboot17:48
eyalrozfsmithred: So now, the apparmor errors are gone, but I get:17:52
eyalroz(evince:3560): dbind-WARNING **: 18:52:27.419: Couldn't register with accessibility bus: Did not receive a reply. Possible causes include: the remote application did not send a reply, the message bus security policy blocked the reply, the reply timeout expired, or the network connection was broken.17:52
fsmithredI'm finding bug reports for that message18:00
fsmithredbut I can't tell which one applies here. Some are old.18:01
fsmithredThese two recommendations keep coming up:18:04
fsmithred1. if the app works and the message isn't too scary, ignore it18:04
fsmithred2. Put the following in /etc/environment18:05
fsmithredexport NO_AT_BRIDGE=118:05
fsmithredeyalroz, I just reproduced your error18:10
fsmithredfirst time I installed evince, I didn't reboot18:10
fsmithredI installed it again and rebooted, and I can't save a file18:11
eyalrozfsmithred: Interesting.18:11
eyalrozfsmithred: Can I help pinpoint the relevant bug report?18:11
fsmithrededit /etc/default/dbus and comment out IDTYPE="RANDOM"18:12
fsmithredand then reboot. Or if you want, I'll test that first.18:12
fsmithredok, that's not enough18:14
fsmithredsorry, I don't think I reproduced the same error. It wouldn't let me save the file because I was trying to save it in its original location, which was in /usr/share/cups18:22
fsmithreddidn't notice that evince did not try to use my home dir18:23
fsmithredso maybe there's a working directory setting in evince18:23
eyalrozfsmithred: Regardless - without apparmor I'm getting a different error, not about permissions;18:24
eyalrozalthough it is significant (for other people) to figure out why that error occurs with apparmor installed18:24
fsmithredyou're right. I would just chalk it up to "it's gnome" and forget about it.18:27
fsmithredCan I interest you in a different pdf reader? Atril is from mate, and it's more like evince used to be.18:28
golinuxI was going to start trashing evince but just didn't have the strength to go there again.18:38
golinuxI dumped it years ago because it sucks.18:38
golinuxAtril is the evince that used to be.18:40
nemodidn't even realise evince was still a thing, but then, have been on MATE for a long time18:44
nemoI did give gnome3 an honest try for 3 months18:44
nemothe shift from windows 7 to windows 10 was pretty painful for me at work.  I mostly resort to using win+x a lot18:45
nemoat least the desktop is default once more18:45
Evilhamsuper+x is pretty much the only way to get to things indeed. but that's OT here :-p18:46
nemoso glad I'm not the only one18:47
nemoEvilham: well the parallel of breaking everything to "help" users is striking18:47
nemoat least microsoft had the excuse that they were hoping to break into mobile18:48
nemowhich is no longer a hope for them, so you'd think they'd rethink some of the tile stuff18:48
nemoif gnome3 was built around dreams of mobile integration they might want to rethink that18:48
nemoesp after death of ubuntu's efforts18:48
Evilhamyeah well, I generally dislike $thing bashing, i fone doens't like something, one tries hard not to use it18:48
nemoEvilham: in microsoft world avoiding it is impossible18:49
nemoEvilham: my poor grandfather is gonna have a rough time when Win7 is EOL'd after decades of windows use18:49
Evilhamwhy I said "one tries" :-p I am aware it's not always possible18:49
nemoEvilham: I'm half considering just shifting him to ubuntu which he has some practice with as a dual boot18:49
nemoaaand maybe a windows 7 VM that's locked down in virtualbox18:49
nemoEvilham: s/ubuntu/devuan/  I'm not sure why I said that18:51
nemoI mean FFS his current laptop is a devuan dual boot, he just rarely uses the linux side ☺18:51
nemoyears of habit I guess18:51
_abc_Hi. Is it possible, in theory, to directly upgrade a machine from say stretch to ascii? If so, please link me to a step by step howto.20:09
furrywolfI'd imagine it's theoretically possible that if you changed your sources.list, did an update, then an apt-get dist-upgrade, it might not break too horribly...20:14
fsmithred_abc_, I did it when ascii was beta or rc20:14
fsmithredand it was the easiest upgrade of all20:14
_abc_fsmithred: good news. So just set the sources etc and fire?20:15
fsmithredum, not quite20:15
fsmithredinstall sysvinit-core (I think that's the right one)20:15
fsmithredthen reboot20:15
fsmithredthen change sources and upgrade20:15
fsmithredalthough I believe others have upgraded first and then switched to sysvinit.20:16
_abc_Hmm? I have to look what is on that machine exactlty. Could be what was "new" before stretch20:16
golinux_abc_: https://devuan.org/os/documentation/dev1fanboy/20:16
fsmithredif it hasn't been upgraded in a long time, there will be a lot of upgrades20:16
fsmithredbut it should still work20:16
_abc_I think it is wheezy, not stretch!20:16
fsmithredomg, upgrade to jessie before July 620:17
_abc_Can one hope for a painless jump from wheezy to ascii? Or is that jumping too far20:17
golinuxThen you should probably go through jessie20:17
eyalrozfsmithred:  Well, I kind of hate evince's UI, because it's Gnome, which I still can't believe nobody has been stoned to to death for...20:17
golinuxYou can keep the pieces.20:17
eyalroz(I was talking about an alternative PDF reader)20:17
eyalrozI'll try atril20:17
golinuxAtril ftw20:17
fsmithredyeah, that one is sane20:17
_abc_golinux: Jessie scrolled out of normal public repos, no?20:18
_abc_Have to use archives?20:18
fsmithredno, there's still jessie20:18
golinuxI'm still using jessie and getting security updates20:18
eyalrozfsmithred: But it seems to have the same problem...20:18
eyalroz(atril:10131): dbind-WARNING **: 21:18:26.610: Couldn't register with accessibility bus: Did not receive a reply. Possible causes include: the remote application did not send a reply, the message bus security policy blocked the reply, the reply timeout expired, or the network connection was broken.20:18
furrywolfis evince the one I tried that doesn't work with icewm?  I don't remember.  heh.20:18
_abc_Ok, good to know.20:18
fsmithredjessie-updates is gone20:18
_abc_Any horrors updating from wheezy to jessie on amd64 known?20:19
fsmithred_abc_, I think it's just minor horrors20:19
fsmithreddon't remember exactly20:19
furrywolfupgrading from wheezy to anything can involve horrors, with adding multi-arch and stuff...  but I don't recall any non-fixable breakage.20:19
golinuxRead the link that has documentation20:19
_abc_Ok. This is a headless machine normally, I hope my KVM still works in case I need it20:20
furrywolfexpect about 100% of your packages to be upgraded.  heh.20:20
fsmithredyeah, dev1fanboy documented it all20:20
_abc_Hm? What link? I missed it?20:20
furrywolfwhy go through jessie instead of ascii?20:20
_abc_ok I see it20:20
_abc_fsmithred: what furrywolf asked? Relevant?20:21
fsmithredwheezy to jessie transition of dependencies has been tested20:21
golinuxhttps://devuan.org/os/documentation/dev1fanboy/20:21
_abc_golinux: yes I found it in the backlog, thanks20:21
golinuxIt's all in there20:21
* golinux goes back to the kitchen20:21
furrywolfwhen you do the apt-get dist-upgrade, pay close attention to what it wants to do...   if it shows it doing stupid things, like removing stuff you either want and/or won't have a functional box without, abort and manually upgrade those things (just apt-get install whatever, and it'll get the newer version) then dist-upgrade again.20:21
_abc_furrywolf: is one not advised to 1st upgrade everything?20:22
fsmithredtry everything first20:23
fsmithredit's also possible you'll need to run dist-upgrade a couple of times20:23
furrywolfif the dist-upgrade looks sane, then just let it run.  but occasionally it does non-sane things.20:23
_abc_I am not sure I have the stamina to watch it doing things.20:23
fsmithredlol20:23
fsmithredit'll be ok20:24
furrywolfapt occasionally does non-sane things in general...  like last week when I tried upgrading wine to the backports version and it wanted to remove >2GB of packages to do it....20:24
fsmithredyou can walk away from it while it's running20:24
_abc_I'll get a snapshot of the important bits I have on there, imaged, and then run it and if it fails I'll do a fresh install.20:24
furrywolfno, you don't have to watch it.  just when it says what will be installed, upgraded, or removed, look at the lists to make sure it's not doing anything stupid, like removing everything.20:24
_abc_fsmithred: how ? ^C?20:24
furrywolfif it's doing something stupid, answer no.20:24
_abc_Ah ok. And then? It skips that step?20:25
fsmithredwalk away one foot in front of the other20:25
_abc_I think I did something like this 2 years ago, once.20:25
fsmithreddownload, unpack and configure can take some time20:25
furrywolfthen it does nothing, and you can try manually upgrading things that it was being stupid about.20:25
furrywolfdo you have a ssd or a hdd?20:25
_abc_Wow sounds like I better install slackware :) hdd and it is old. But I have time. The problem is the din. amd64 boxes are not delicate about fans and so on.20:26
furrywolfif you have a hdd, the walk away step takes about ten times longer.  :)20:26
_abc_I should buy a ssd finally. I assume it is much faster.20:26
furrywolfI couldn't believe how much faster apt worked once I got a ssd.20:26
fsmithredyou're doing this over ssh?20:26
_abc_I think the people who wrote apt "think" in terms of Visual Basic 9th grade kid efficiency. I have no other explanation for it's sloth.20:27
nemoI have a 4 terabyte HDD20:27
_abc_fsmithred: Assuming I can squeeze in a $30 new 120GB ssd and image the partition onto it before starting.20:27
nemoI don't see any purpose in an SSD20:27
_abc_nemo: speed20:27
nemoin terms of lifespan they are roughly equivalent - esp for the always-up model20:27
_abc_nemo: for when script kiddies code important things like apt20:28
fsmithredI got a 240GB laptop ssd for $3020:28
furrywolfmuch, much, much faster.  this is a purpose.20:28
nemo_abc_: boot speed is irrelevant. application launch rarely matters. since most apps I routinely launch are cached in ram anyway20:28
_abc_We are in backwaters here, a low end Kingston 120GB sata ssd is about $3020:28
fsmithrednewegg20:28
fsmithredor maybe you can't get there20:29
nemothe 4 terabyte HDD cost me $10020:29
nemoI'm sure solid state will beat HDs soon, but, eh, I'm in no rush20:29
_abc_fsmithred: I can but it's a hassle. The Kingston is in a walk in shop about 15 minutes walking away.20:29
furrywolfit's not just launch speed...  it's everything.  programs are always creating temp files and such, and everything is amazingly stupidly faster with a ssd.20:29
fsmithredanyway, if it's headless, and it fails and you can't get in, you can boot a refracta iso and ssh in as user.20:29
nemoI have an SSD on the laptop and I don't notice any difference in speed20:29
nemofurrywolf: temp files and such in linux are usually created in /tmp20:29
nemowhich is rambacked by default20:29
nemofurrywolf: I really don't notice any difference20:29
nemomaybe windows it matters more20:30
_abc_fsmithred: yes, that is the plan. I am just worried I'll have to reinstall and license all the crap I have on there frozen about 2 years ago.20:30
furrywolfnemo:  you can argue theory all you want, but in reality, every box I've stuck a ssd in has immediately been much faster in every normal-use-case way.20:30
nemofurrywolf: I do have 16 gigs of ram on the machine, but that's nothing special by today's standards20:30
nemofurrywolf: heh. I'm not arguing theory. I'm arguing reality of my two machines here and how I use them20:30
furrywolffirefox creates a billion or so tiny cache files, all of which it loads faster every time you load a web page.  etc.20:30
furrywolfevery program that uses sqlite for data functions much faster20:31
furrywolfetc etc20:31
fsmithredlicense? crap? If you have outside programs installed, are you sure they'll work after the upgrade? Or will they be upgraded, too?20:31
_abc_furrywolf: we know you are right, nemo is hanging on to tradition. In reality, with good management, one could have both a spinning hdd and a ssd in the same machine. Which is what I will have when I get a ssd/ if I get it.20:31
_abc_fsmithred: they are not so dependent on the version, fpga tools, vivado, eclipse, mplab x and so on20:31
_abc_netbeans ide . I wonder what I keep that one around for now.20:32
nemofurrywolf: if your sqlite file is frequently accessed it will be cached too20:32
furrywolfnothing that needs a license is included in the debian repos.  none of that will be touched.20:32
nemofurrywolf: I currently have 10½ gigs of cached files apparently20:32
nemomy firefox profile is tiny compared to that20:32
_abc_furrywolf: the problem is the licensed tools look at the os for clues on whether they're still on the machine they are intended to be on20:32
fsmithredI have the new ssd in a box along with an old spinning drive and the difference is dramatic. But I don't think it would be much different from the sata III spinner that's in the laptop.20:32
nemofurrywolf: in terms of firefox disc cache, yes, that will be faster, ofc we're comparing to network fetches by default20:32
nemoaaaand it will be ram cached too20:33
_abc_fsmithred: difference measured how? Real life read write or benchmark or...20:33
furrywolfsqlite by default insists on flushing everything before the calls return.20:33
nemo_abc_: I wouldn't hang on to tradition if SSD wasn't so stupid-expensive ☺20:33
_abc_F.ex. a compilation or the kernel, would it go faster? fsmithred20:33
_abc_nemo: low end ssd is now cheaper than low end spinning drives20:34
nemo_abc_: I could certainly do SSD + HD - might do that for next machine20:34
fsmithreddifference measured by perception. I'd need to look at my watch to compare the two laptop drives.20:34
nemo_abc_: move all the giant files onto the HD20:34
furrywolfsqlite won't tell the caller that an insert/etc is done until it's on physical media.20:34
nemo_abc_: on current machine there is no place for it. in part due to a stunningly stupid motherboard fail20:34
fsmithreddon't need any timer to notice the difference between the old drive and the new ssd.20:34
nemoI reboot desktop maybe once every few months. it might be more noticeable on laptop, although I don't reboot that one very often either20:35
nemoit is pretty annoying though that I'm already at 90% usage of my laptop HD ☹20:35
nemoand this is after deleting a ton of stuff20:35
nemo/dev/sda1       219G  186G   21G  90% /20:35
nemo↑ laptop SSD20:36
nemo/dev/sdb3             3.6T  2.0T  1.5T  58% /20:36
furrywolfthat's a lot of porn.20:36
nemo↑ desktop HD20:36
nemofurrywolf: heh. laptop is a work computer. no porn20:36
nemofurrywolf: they can take it and image it at any time, setting aside their network monitoring20:36
nemoit's mostly a ton of development files and images and libraries20:36
nemoimages as in VMs20:36
nemoVMs for testing windows and android...20:37
_abc_nemo: what's the hdparam -tT reported speed on that 4T drive?20:38
nemogonna guess you meant to say hdparm...20:42
nemo Timing cached reads:   7600 MB in  2.00 seconds = 3802.85 MB/sec20:43
nemo Timing buffered disk reads: 346 MB in  3.01 seconds = 114.79 MB/sec20:43
nemoby comparison the work SSD is...20:43
nemo Timing cached reads:   12336 MB in  2.00 seconds = 6179.56 MB/sec20:43
nemo Timing buffered disk reads: 1490 MB in  3.00 seconds = 496.09 MB/sec20:43
nemooverall... not significant enough to make a difference most of the time. even in the scenarios he mentioned20:44
furrywolfdo a random 4k read test.20:44
furrywolfthe difference is like 1000 times.20:45
furrywolfand is much more represntative of normal application usage20:45
nemowhy would random be more representative of normal?20:45
nemotrying hard to imagine what app would have a usage case like that20:45
nemonot to mention, again, small recently accessed files are far more likely to be cached in ram, even on first write.20:46
furrywolffirefox does exactly that.20:46
_abc_In the sense of https://devuan.org/os/documentation/dev1fanboy/general-information -- what is an "embedded" .xz image? I think I never saw one. Or did and did not pay attention? Where would such an embedded image be found?20:46
furrywolfthousands of few-kb files randomly loaded20:46
nemoonly 1000 out of the 11000 files in my firefox cache were loaded today. and I did a LOT of browsing20:47
nemototal size, 100MB20:47
nemotrivially ram backed20:47
nemos/loaded/written/20:47
* _abc_ remembers when he was using wwwoffle 20 years ago as cache with a modem.20:48
furrywolftime an apt-get upgrade, which is what started this, on both a hdd and a ssd.  :P20:48
furrywolfapt is another thousands-of-tiny-files operation.20:48
nemoI'm not disputing that operations that involve a lot of updating of infrequently accessed stuff will be slower20:49
nemoin terms of my desktop experience though, that's basically irrelevant20:49
nemofrankly I don't really care how long the upgrade takes 99% of the time20:49
nemohell, I have gentoo systems so I must *really* not care 😉20:49
furrywolfin my experience, normal desktop usage is vastly better with a ssd.  to the point where I'm never using a hdd for anything other than backups again.20:50
nemonow if the upgrade forced me off the desktop while it was upgrading, that sure would suck20:50
nemoand in my experience exactly opposite! 😃20:50
nemoso lucky for me I'm happy eh20:50
* nemo hugs his 4T drive20:50
nemowhich cost him $10020:50
nemohm20:50
nemowonder how much a 4T SSD would cost20:50
furrywolf$45020:50
stiltrIt's interesting how many fellow gentoo people are here in #devuan. :)20:51
_abc_devuan is systemd less gentoo? :)20:51
* _abc_ would be interested in the expertise of arch people20:51
nemohttps://www.newegg.com/p/pl?d=4T+SSD&N=-1  that's clearly on the low-end. but yeah, no thanks20:52
nemo_abc_: systemd is optional in gentoo20:52
nemo_abc_: hell. on a couple of my gentoo systems I run a complete poetering purge20:52
stiltrEverything is optional in gentoo. haha20:52
_abc_So, again, what would an "embedded" image be?20:52
nemojust 'cause it is easier to pull off20:52
nemohttps://m8y.org/tmp/package_mask.txt20:52
_abc_gentoo is the one where you compile everything from scratch? Or is that arch?20:52
nemostiltr: yeah, but they do go to the effort of maintaining alternatives. which is nice of them20:53
nemostiltr: they were one of the first ones actually...20:53
stiltrFor sure!20:53
stiltr_abc_: I think arch is basically gentoo with binary packages.20:54
nemohttps://wiki.gentoo.org/wiki/Hard_dependencies_on_systemd20:54
nemostiltr: gentoo always had binary packages but they are very poorly maintained these days20:54
nemomaybe due to arch. dunno20:54
nemoyou can ofc maintain your own binary repo locally20:54
stiltrYa, you can do bin packages in gentoo, but I think arch is bin-only. That said, I have very limited experience with arch.20:55
nemohm. only 6 packages with hard depends on systemd. not bad20:56
nemosetting aside systemd itself naturally I imagine ☺20:56
nemomutter and gnome is no shock20:56
stiltrYa, I was pleasantly surprised by that.20:56
nemohm. and one of the others is only "kinda requires systemd"20:56
_abc_A sign that the cancer has not spread to the bones yet20:57
* _abc_ heads for #slackware to ask about systemd penetration there20:57
fsmithred_abc_, embedded images are disk images (.img) not isos. You just dd the system to the device. (e.g. raspi)20:58
fsmithredthe .xz is compression20:58
_abc_Ah those. Non pc platform "firmware".21:00
_abc_I will hate google forever for letting android call the firmware flash image "rom"21:01
fsmithredrofu? (read-only-for-user)21:02
stiltrI'm not sure I'd call it firmware in the case of a full OS, but ya pretty much. Definitely agree on the misuse of ROM.21:02
stiltrhaha21:03
fsmithredno user serviceable parts inside21:03
fsmithredor bits21:03
stiltrI always take that more as a challenge...21:04
_abc_yes, pity cyanogenmod became a tame app supplier lately.21:06
fsmithredso the ssd is more than 4x the speed of the old drive on buffered disk reads21:18
fsmithredabout 10% faster on cached reads21:19
_abc_Write is not slower, right? On ssd vs hdd? fsmithred?21:20
furrywolfssds are much faster, especially at random writes.21:21
_abc_Sure, no head seek time.21:22
_abc_But sustained write speed? Faster too?21:22
fsmithredI only tested reads21:22
furrywolfyes, sustained write is much faster too.21:24
furrywolfanything remotely modern is limited by the sata iii interface, which is why nvme became a thing.21:24
furrywolfheh, and with shingled recording, modern hdd write speeds are actually going back down in some cases.21:26
furrywolfthe seagate backup drives I got are disgustingly slow.21:26
_abc_Shingled slows down? Interesting.21:28
furrywolfincredibly much so21:28
furrywolfall the tests that claim acceptable write speeds are either small tests where the drive writes only to its non-shingled buffer areas, or large sequential writes (>32MB)...21:29
furrywolfhttps://www.storagereview.com/seagate_archive_hdd_review_8tb  has a bunch of tests.   https://www.storagereview.com/images/seagate_archive_8tb_sata_main_4kwrite_throughput.png  sums up my experience with mine quite well...21:30
furrywolfone of those bars is not like the others.21:30
furrywolflike a ssd, a smr drive can only write large blocks in one go.  to change a single sector, it has to read the entire block (32MB for those drives if I remember right) to ram and then write the whole block back.21:31
_abc_Yes I am aware.21:32
_abc_Just did not think about the implications till now.21:32
_abc_Yeah the archive drive clearly is not happy writing there21:33
furrywolfto make this acceptable, they use a non-shingled buffer area.  as long as you don't write more than that buffer area can hold, the drive will use its idle periods to slowly move that data out of the buffer and to where it should go.  but if you exceed the buffer, it has to do a full read-modify-write operation for every write...  which takes a third of a second apparently.21:33
_abc_furrywolf: I assume also it does not like unscheduled power downs while doing this at all21:34
furrywolfI would assume that too, but I didn't see any tests of that when I was reading about them.21:34
_abc_Is there some minimal guarantee that data loss will not occur in case of unscheduled power down after the host wrote the buffers but while the drive is shoeshining that block?21:34
furrywolfhttps://www.storagereview.com/images/seagate_archive_8tb_sata_main_8k7030_throughput.png  even with larger blocks and both reads and writes mixed...  no, that's not the X axis there at the bottom...21:35
_abc_I get it.21:36
_abc_Well, is the price at least proportionally lower? ...21:36
furrywolfI bought a seagate backup drive at costco, and was trying to figure out why the fuck it was stupid pathetically god-awfully nonfunctionally slow.  at first I thought the drive was defective...  then I googled and learned about these things.21:37
_abc_Aren't these the Helium filled ones? Or is that another thing?21:37
furrywolfthey're pretty cheap, yes.  I wouldn't buy one again.  not worth it.21:37
furrywolfI have two of them.  really only good for backups, and ones that you don't need to sit there for.21:38
_abc_Interesting ambient does not give the usual max. operating altitude. I read that as them being pressurized.21:38
furrywolfI don't have the exact version in that article...  I have the 4tb 2.5" version.21:40
furrywolfdamnit, I have to go to work soon, and that means putting a top on.  it is way too fucking hot today.21:41
_abc_Well they give 1/1e15 read probability error, that is low enough.21:41
_abc_HGST is who? Hitachi Toshiba ?21:43
furrywolfhitachi, ibm, wd...21:45
_abc_https://en.wikipedia.org/wiki/HGST ah21:46
furrywolfgoogling says seagate at one point was saying they used the energy of the platters spinning the spindle motor to power the drive long enough to orderly shut down.21:49
furrywolfthey keep pointing at the original until the new data is completely written, so it only needs to power itself long enough to update the metadata21:50
furrywolfif the big block write is interrupted, nothing bad happens, since it's not being pointed at yet21:50
furrywolfand the metadata update is quick enough it has enough stored energy to finish it21:51
_abc_Makes sense.21:53
furrywolfthe ssds in the boxes I care about are enterprise ones, which have a big bank of capacitors to make sure they can finish their housekeeping...  using the kinetic energy of the platters for the same purpose on a hdd is clever.21:55
furrywolfbbl, time to put on socially-acceptable levels of clothing and head to work.21:57
_abc_It is also old. As an idea. Backup online for big machines was ensured by the rotary converter spinning down, while still generating, until the Diesels started.21:57
_abc_It also bridged any gap while relays switched over to battery powered rotary generator etc.21:57

Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!