|
PIGEOTO posted:More ram, haha
|
# ? Oct 28, 2014 11:14 |
|
|
# ? Apr 26, 2024 19:25 |
|
Notorious b.s.d. posted:fuckin lol at 4 gb in a developer desktop
|
# ? Oct 28, 2014 13:06 |
|
Notorious b.s.d. posted:incidentally this is why you monitor swapin/swapout rates and not swap usage directly That is why I didn't pester Red Hat. Sar showed no actual swapping. It just sat there, unused, bugging the poo poo out of me every time I manually ran free. RHEL 3. The good (bad) old days. pram posted:vm.swappiness you moron Cool.
|
# ? Oct 28, 2014 13:07 |
|
keyvin posted:That is why I didn't pester Red Hat. Sar showed no actual swapping. It just sat there, unused, bugging the poo poo out of me every time I manually ran free. so you had no sign whatsoever of a problem you just wanted to sperg out about a number in tool output
|
# ? Oct 28, 2014 15:07 |
|
Notorious b.s.d. posted:so you had no sign whatsoever of a problem in a thread where we all sperg about words in tool output (its u)
|
# ? Oct 28, 2014 15:22 |
|
Cocoa Crispies posted:in a thread where we all sperg about words in tool output (its u) lol
|
# ? Oct 28, 2014 16:17 |
|
lot of server janitoring going on in this desktop thread
|
# ? Oct 28, 2014 16:33 |
|
Notorious b.s.d. posted:so you had no sign whatsoever of a problem i live in a world where operations guys get paged immediately if a server hits swap because that is an error condition - Notorious b.s.d.
|
# ? Oct 28, 2014 16:56 |
|
ZShakespeare posted:lot of server janitoring going on in this desktop thread Asperger's Syndrome FTW
|
# ? Oct 28, 2014 17:26 |
|
keyvin posted:i live in a world where operations guys get paged immediately if a server hits swap as mentioned you should be monitoring swap rate
|
# ? Oct 28, 2014 17:40 |
|
Handlebarred ad agency Mac-guy 'creatives' were insufferable enough, and now they're being joined by the worst BSD-lords. Between this and HTML5/Node.js buttholes, I don't wanna work in computers anymore.
|
# ? Oct 28, 2014 19:32 |
|
Captain Pike posted:Handlebarred ad agency Mac-guy 'creatives' were insufferable enough, and now they're being joined by the worst BSD-lords. Between this and HTML5/Node.js buttholes, I don't wanna work in computers anymore. lol at wanting to work in computers in the first place
|
# ? Oct 28, 2014 19:50 |
|
bsd owns (as in: osx, not actually bsd which is dead (netcraft confirms))
|
# ? Oct 28, 2014 19:53 |
|
osx is the worst bsd
|
# ? Oct 28, 2014 19:58 |
|
OldAlias posted:osx is the worst bsd lol
|
# ? Oct 28, 2014 19:59 |
|
all bsd is pretty terrible in general sysv supremacy
|
# ? Oct 28, 2014 20:06 |
|
Notorious b.s.d. posted:all bsd is pretty terrible in general
|
# ? Oct 28, 2014 20:08 |
|
Captain Pike posted:Handlebarred ad agency Mac-guy 'creatives' were insufferable enough, and now they're being joined by the worst BSD-lords. Between this and HTML5/Node.js buttholes, I don't wanna work in computers anymore. come join us in embedded systems! we have bad compilers and worse hardware! thursday night is open mic
|
# ? Oct 28, 2014 20:08 |
|
Please rename this thread to swapfile.avi
|
# ? Oct 28, 2014 20:32 |
|
Fib posted:Please rename this thread to swapfile.avi
|
# ? Oct 28, 2014 21:37 |
Soricidus posted:that would also be a good yospos username i'm reasonably sure ozma gave that to someone last year
|
|
# ? Oct 28, 2014 21:49 |
|
keyvin posted:i live in a world where operations guys get paged immediately if a server hits swap that's a really big fuckin problem your app running on the jvm is likely making GBS threads itself at this point and either you handle it now or you handle it 3am when it goes hard down because you're infinitely GCing
|
# ? Oct 28, 2014 22:21 |
|
wish ur posts would be gc'd
|
# ? Oct 28, 2014 22:25 |
|
pram posted:wish ur posts would be gc'd
|
# ? Oct 28, 2014 22:26 |
|
Back to desktop computing: btrfs, the hot new filesystem that takes journaling to the extreme, doesn't support swap files (but you can create a loop device on a file and use that, but I don't think that would work with kdump). With the rate of btrfs adoption in default desktop installs, I don't think we're gonna get dynamic swapfiles anytime soon. Personally, I think just having one or two gigs of swap space and reacting if the desktop is dog slow is the best solution.
|
# ? Oct 28, 2014 22:57 |
kittenkicker posted:and reacting if the desktop is dog slow format install windows
|
|
# ? Oct 28, 2014 23:06 |
|
a friend just let me know he installed Fedora 21 on a mbp retina. he said it works really loving well. i have asked no follow up questions. another linux laptop success. 2015 is looking good as year of the linux on the laptop.
|
# ? Oct 28, 2014 23:45 |
|
congratulate him on installing the best os on the best laptop
|
# ? Oct 28, 2014 23:46 |
pram posted:wish ur posts would be gc'd
|
|
# ? Oct 29, 2014 00:19 |
|
kittenkicker posted:btrfs, the hot new filesystem that takes journaling to the extreme, doesn't support swap files (but you can create a loop device on a file and use that, but I don't think that would work with kdump). With the rate of btrfs adoption in default desktop installs, I don't think we're gonna get dynamic swapfiles anytime soon. yeah that 0% uptake rate is really gonna change the direction of linux
|
# ? Oct 29, 2014 02:04 |
|
Kiwi Ghost Chips posted:congratulate him on installing the best os on the best laptop
|
# ? Oct 29, 2014 02:04 |
|
i don't care much for btrfs actually it's been in development for like seven years and it's still not considered stable it merges a whole bunch of layers into one soupy mess, which seem like they could be solved just as effectively using discard-aware lvm and ext4 elaborate and fragile on-disk data structures with a shitton of invariants that absolutely have to hold all for the purpose of minimising seeking aren't a worthwhile tradeoff in the age of ssds. for server applications with big rotating platter hdds you'll have a database type thing that carves out a massive chunk of disk and manages it in an application-specific way anyway as a matter of fact i'm shifting over to thinking that runit is a better way of accomplishing what systemd is trying to do. socket activation is nice but it's an orthogonal problem: have runit start up a bunch of stub processes that listen on unix sockets or tcp sockets or dbus busnames or w/e and then fork and handoff to a real server process when those sockets first get opened. launch that entire tree with its stdout redirected to a logger. for each service you have one process for supervision, one for logging, one for socket activation, and one to actually do the work. this is unix, processes are cheap. no big elaborate pid1 with a multitude of responsibilities that can take the entire system down if it segfaults necessary. controlling services using dbus is kinda nice i guess? but runit uses single-ascii-byte commands to named pipes to control its supervised services and that seems to work well enough. as a bonus you can actually use filesystem permissions to control access to them too.
|
# ? Oct 29, 2014 02:29 |
|
runit owns lvm owns ext4 is p. nice xfs owns i honestly do not understand the btrfs project goals when we live in a world where lvm and xfs already exist it's telling that after all these years of btrfs development no one, not even the weird eurotrash @ suse, have picked it up
|
# ? Oct 29, 2014 02:33 |
|
Mr Dog posted:as a matter of fact i'm shifting over to thinking that runit is a better way of accomplishing what systemd is trying to do. socket activation is nice but it's an orthogonal problem: have runit start up a bunch of stub processes that listen on unix sockets or tcp sockets or dbus busnames or w/e and then fork and handoff to a real server process when those sockets first get opened. launch that entire tree with its stdout redirected to a logger. for each service you have one process for supervision, one for logging, one for socket activation, and one to actually do the work. this is unix, processes are cheap. no big elaborate pid1 with a multitude of responsibilities that can take the entire system down if it segfaults necessary. controlling services using dbus is kinda nice i guess? but runit uses single-ascii-byte commands to named pipes to control its supervised services and that seems to work well enough. as a bonus you can actually use filesystem permissions to control access to them too. lol. if upstart, backed by canonical, already lost then what makes you think fuckin RUNIT has any chance
|
# ? Oct 29, 2014 02:35 |
|
well, the reason that systemd is so popular is because it actually standardises things and does useful things out of the box that the other system service managers on linux can eh maybe kinda sorta do if you screw around with them long and hard enough and make extensive modifications from upstream that'll only be relevant for your distribution the greatest thing by far about systemd is that it made distributions almost completely irrelevant at a stroke. literally the only difference between distributions now is who you sign the support contract with, the package management system that it uses, and the release cycle. systemd could be Satan itself and the foregoing would still make it a net positive for linux. i think that's a large part of why people hate systemd (not dislike or aversion, i do specifically mean hate: it's an intensely emotional reaction that is retroactively rationalised as opposed to actually being a considered objection based on engineering principles) well, actually i'll amend that last statement: it could be much much worse, it could be the Windows Event Log (which requires you to compile "message catalogs" of i18nable format strings using a tool that hasn't been updated since 1985, in UTF-16 of course, and then refer to those messages by number from your source code any time you wanted to log something), it could have the NT Service API and provide a libadvapi32 that you had to link against and write a shitton of boilerplate for if you wanted to run as a service which would also make it impossible to run said process outside of the Service Manager, and you could have a binary registry tracking both configuration and state as the cherry on this poo poo sundae. I think that's another reason why people hate systemd: they probably just assume that it faithfully replicates the horror show that is the NT service API instead of providing an interface that isn't designed by a sadistic psychopath. anyway that kind of went off on a tangent. systemd standardises things and solves real problems. but the manner in which it solves those problems is suspect: i think systemd's pid1 could actually be quite elegantly split into a set of orthogonal co-operating executables and still achieve just about everything the current solution a achieves. the systemd guys do give an outline for why the linux cgroup tree needs to be managed by a global supervisory process with a complete view of the entire system's operation, but this actually seems to be the crux of the argument for a monolithic pid1 and in light of that the whole document is a little too imprecise, hand-wavy, and short on concrete examples for my liking. Sapozhnik fucked around with this message at 02:50 on Oct 29, 2014 |
# ? Oct 29, 2014 02:48 |
Mr Dog posted:it's an intensely emotional reaction that is retroactively rationalised as opposed to actually being
|
|
# ? Oct 29, 2014 02:49 |
|
systemd's PID1 isn't monolithic and systemd is a set of co-operating executables (I restart journald and logind all the time during development, without taking down PID1).
|
# ? Oct 29, 2014 02:50 |
|
Mr Dog posted:the greatest thing by far about systemd is that it made distributions almost completely irrelevant at a stroke. literally the only difference between distributions now is who you sign the support contract with, the package management system that it uses, and the release cycle. systemd could be Satan itself and the foregoing would still make it a net positive for linux. i think that's a large part of why people hate systemd (not dislike or aversion, i do specifically mean hate: it's an intensely emotional reaction that is retroactively rationalised as opposed to actually being a considered objection based on engineering principles) That's the "Microsoft-ification of Linux" argument: that we should probably have an enterprise OS that we all sort of agree on, rather than a giant bucket of random components built by one vendor or another. I think it's a great thing to finally have platform to target and one API to use. Others have stronger emotions against that.
|
# ? Oct 29, 2014 02:52 |
|
Suspicious Dish posted:That's the "Microsoft-ification of Linux" argument: that we should probably have an enterprise OS that we all sort of agree on, rather than a giant bucket of random components built by one vendor or another. I think it's a great thing to finally have platform to target and one API to use. Others have stronger emotions against that. i'm totally on board just the one OS and vendor should have been sun/solaris
|
# ? Oct 29, 2014 02:55 |
|
|
# ? Apr 26, 2024 19:25 |
|
Mr Dog posted:i don't care much for btrfs actually iirc it was originally supposed to be the linux zfs or something but now there's a native zfs for people that actually need it
|
# ? Oct 29, 2014 02:59 |