I Moved to Linux and It’s Even Better Than I Expected
-
It makes sense that accessing a partition formatted as swap directly incurs less overhead than accessing a file formatted as swap through a filesystem layer.
But the maker of the OS is in control of the filesystem, so... I don't see that as some kind of truism.
Unlikely; once a swap file is initially opened, the kernel knows where all its disk blocks are and no longer has to scruffle though filesystem code to find them.
^- yeah, that
-
How do other OSes do it? That exact feature works better in Windows and OS X than it does in Linux.
Both of those integrate fairly tightly with the BIOS or UEFI, and when the machine comes out of suspend or hibernation it's typically not actually rebooted. Linux's suspend-to-disk/hibernate stuff works on any architecture regardless of firmware support; the initial boot process is completely normal, and only once a kernel is up and running is a suspended image searched for and reloaded. I believe Windows can do this too, if the machine has been fully powered off between hibernate and resume; Linux does it every time.
-
Don't see the advantage over just copying the directory over.
I have, or had for a while, both distros installed, sharing the same home directory.
Wanna hear how you could change Mac Classic versions without even formatting the disk? Up and downwards?
That's a cool feature, but not relevant as I am using linux distros, and Mac Classic won't even run on my computer.
-
It makes sense that accessing a partition formatted as swap directly incurs less overhead than accessing a file formatted as swap through a filesystem layer.It's one level of indirection less to go through.
It would make sense if that's what happened, but that's not what happens. Once the swap file is opened, the kernel knows where all its disk blocks are and accesses those directly. Swap files get locked so that the underlying filesystem can't move stuff around while it's in use.
The only performance difference would be down to extra seeks caused by disk fragmentation. Fragmentation is typically very low (under 2%) in most Linux filesystems, so it's not a big factor. But really, if you care about swap file performance you're
in any case.
-
running several different OS-s and having them share a home partition is a pretty niche use case. Also sounds like a recepe for disaster if software versions drift too far apart.
Bit of a pain when there's more than one user on the box as well; you have to make sure that all your installed OSes agree about which UID is which.
-
@OffByOne said:
It makes sense that accessing a partition formatted as swap directly incurs less overhead than accessing a file formatted as swap through a filesystem layer.
But the maker of the OS is in control of the filesystem, so... I don't see that as some kind of truism.
? The maker of the OS provides drivers for the various filesystems it supports. A swap file on a filesystem is "just" a regular file that happens to be formatted in such a way that it can be used by the kernel as virtual memory.
Access to that file happens through the filesystem driver for the filesystem the swap file lives on. Since different filesystems have different layouts on disk, the virtual memory subsystem can't make any assumptions about the physical layout of the swap file on disk, right?@flabdablet said:
Unlikely; once a swap file is initially opened, the kernel knows where all its disk blocks are and no longer has to scruffle though filesystem code to find them.
^- yeah, that
That could be a special code path for files used as swap space to shortcut the filesystem driver, but I wouldn't assume that works for all filesystems.
What if your swap file is on a RAID filesystem? What if your swap file is on an encrypted filesystem? The virtual memory subsystem can't just access raw disk blocks, it needs to go through the proper filesystem driver.
-
Unlike Windows, there's few things you can outright dowload and run on linux, even if you have root.
There's masses of stuff in the repos. But there are certainly way, way, WAY fewer random setup.exes for ill-judged impulse installs, and for n00bs that's actually huge.
You can't rely on a n00b to understand that e.g. Uniblue Registry Optimizer Pro is not a good thing to stick on their Windows box. Easily the most common cause of support calls from Windows customers is cleaning up after they, or actually more typically their kids, have installed some piece of oooh look shiny! that they had no idea was actually some kind of Trojan. Video and music downloaders are a pretty common vector.
Linux customers, on the other hand, will typically try to download and run a few of these things, get nowhere because they're not compatible, and then either give up or call me.
-
Once the swap file is opened, the kernel knows where all its disk blocks are and accesses those directly.
Citation needed. What happens in the "swap file on an encrypted filesystem" scenario I described in my previous post?
The only performance difference would be down to extra seeks caused by disk fragmentation. Fragmentation is typically very low (under 2%) in most Linux filesystems, so it's not a big factor.
Sure, it's not a big factor, but Blakey's question was "is it measurable"? I'm too lazy to find actual numbers, so my answer to that question is "I don't know for sure, but my intuition is telling me it's not zero. It might be negligible for most practical purposes, it might not be.".
But really, if you care about swap file performance you're
in any case.I'm using Discourse. Of course I care about swap file performance! I do agree with you in reality though.
-
I don't know why it needs a separate partition.
Especially since you can create swap memory in a file too, like on Windows.
IIRC a long time ago swap had to be on a separate partition, so like someone upthread said, anyone telling you to do it today is just echoing ancient advice.
-
Noob running linux is like balancing on a rope over a chasm. Everything is cool as long as you stay within your safe tested workflows and don't look down.
Worst yet, he's using it as an environment for his clients. And from the list, looks like lightweight highly customized software at that.
When he's gone, his clients are going to hate him.... deeply.
-
Since different filesystems have different layouts on disk, the virtual memory subsystem can't make any assumptions about the physical layout of the swap file on disk, right?
-
What happens in the "swap file on an encrypted filesystem" scenario I described in my previous post?
The kernel bmaps the file blocks, then accesses them via the block device driver.
If I recall correctly, Linux will ignore swap files declared on filesystems that can't be bmapped.
-
Thanks for the information I was too lazy to look up myself.
I'll read it tomorrow, it's getting late here and the fact that I'm enjoying my Christmas present (Talisker Storm) doesn't really help my reading comprehension at the moment :)
-
Since the GRUB 2 boot loader has existed (at least ten years now) there's no longer any good reason to use a separate /boot partition on x86 boxes.
Software RAID? LVM? LUKS? Does grub support all of that now? Last time I checked, it couldn't handle more complex setups. (I use all 3.)
-
How do other OSes do it? That exact feature works better in Windows and OS X than it does in Linux.
Yeah, sadly. I tried to use suspend-to-disk* for a while… I got it to work multiple times, but it broke with every upgrade. After a while, I just gave up. I mostly blame shitty drivers/BIOSes, but I've heard that there are other problems involved as well.
*Abbreviated STD for a reason.
-
GRUB 2 has modules for all three of those, and I believe it does let you stack them in any order.
The main part of GRUB 2 (the one built from all the modules) will typically be rather bigger than GRUB 1's stage 1.5, and if you need lots of disk access modules it will often fail to fit in the unpartitioned space on any disk formatted old-skool where partition 1 starts on LBA 63. On MBR-partitioned disks where partition 1 starts on LBA 2048, which has been pretty common since Windows Vista, never any problem at all.
On GPT-partitioned disks that are likely ever to be booted in legacy BIOS mode, I like to tuck in a little GRUB partition (with the bios_grub flag set using parted) at LBA 128..2047; that puts it before the place where the first partition would otherwise start, so it costs literally zero space.
-
GRUB 2 has modules for all three of those, and I believe it does let you stack them in any order.
Exactly what I highlighted above didn't work the last time I checked.
-
I mostly blame shitty drivers/BIOSes
Linux suspend-to-disk doesn't rely on BIOS support AFAIK. Suspend-to-RAM does. Most of the issues with these things failing do come down to driver misbehavior, often due to firmware issues.
-
didn't work the last time I checked.
What distro? Debian's pretty good at making all this stuff Just Work.
-
Linux suspend-to-disk doesn't rely on BIOS support AFAIK.
I really don't know much about that kind of stuff, but I think I read that it relies on ACPI tables that the BIOS provides, and that those are often broken.
-
What distro? Debian's pretty good at making all this stuff Just Work.
Ubuntu. Also not the distro's fault IIRC, I think I subscribed to a related grub2 bug report at the time.
-
Just a thought: if you're using LUKS to encrypt an entire raw disk, rather than encrypting partitions on that disk, then I don't believe GRUB will be able to boot off that; the crypto read stuff is a module, and that puts it in GRUB's second stage, and if you've put that second stage into a partition created on top of LUKS then the first stage isn't going to be able to find it.
Fix would be to leave the partition table plus the space that GRUB's second stage will go in (unpartitioned space following the MBR if the disk is MBR formatted, or a <1MB partition with the bios_grub flag set if GPT formatted) unencrypted; then make a big partition covering the rest of the disk and LUKS-encrypt that.
Edit: that's the method used here.
There are no secrets embedded in GRUB's second stage, so you wouldn't be exposing anything by doing it that way. Your disk would, however, still be vulnerable to having GRUB's second stage tampered with. Only way I can think of to avoid that would be to use a crypto-aware UEFI boot sequence - if you're running second-stage GRUB as a UEFI boot loader (on Debian this would require installing the grub-efi package rather than the grub-pc one) and the EFI system partition is itself encrypted, then that should tamper-proof the entire boot chain.
-
IIRC (the laptop died last month), I tried RAID 1 -> LVM -> LUKS -> root partition (no extra /boot) and that didn't work for some reason.
-
As far as I know, that configuration is now supported without needing to do anything outrageously clever.
-
Having /home separate or on a file share allows easier multibooting, or reusing the home partition on multiple hosts
It also means you can blow away the system partition without trashing all your other stuff if you need to do a complete reinstall. Certain distributions (like RedHat and Fedora) really want you to do this when upgrading to a new major-version-number release.
(Yes, it's dumb. And yes, I've ignored that advise and successfully upgraded. But that's still what they recommend.)
I'm not sure on the reasoning for having /usr/local separate
I don't know what the official reason is, but my guess is that it's for the same reason you'd keep /home separate. Stuff in /usr/local is generally stuff you compiled yourself (content bundled in packages typically installs under /usr or some other similar location.) It's very convenient to not have to recompile and reinstall everything if you need to nuke-and-pave the OS.
Of course, a major upgrade may force you to recompile that code anyway, due to shared library incompatibilities, but not necessarily everything.
I'm not sure off the back of my head of the exact advantages of a swap partition over a swap file nowadays,
I don't think it matters too much today. Originally, it would improve performance because the swap could never become fragmented, having a partition all to itself. But with large hard drives and large caches (both in hardware and software), the benefit is not what it used to be. And if you're storing your swap on an SSD, there may be no significant benefit at all.
-
When problems occur, the communities that have emerged around free and open-source software are incredibly helpful.
HahahahhahAHHAHAHAHAHAHAHHAHAH
:lau
edit: what the shit, Discourse? Pressing tab doesn't autocomplete the emoji selector, and enter randomly posts now.
Anyways:
-
? The maker of the OS provides drivers for the various filesystems it supports.
I believe I just said that.
Since different filesystems have different layouts on disk, the virtual memory subsystem can't make any assumptions about the physical layout of the swap file on disk, right?
Well since the maker of the OS is in control of the filesystem, yes, they could easily put a "mode" in the filesystem that does whatever thing having the swap file on a different partition does to make it faster. Whatever that is. Assuming it's even faster anyway. Which I doubt.
What if your swap file is on a RAID filesystem? What if your swap file is on an encrypted filesystem? The virtual memory subsystem can't just access raw disk blocks, it needs to go through the proper filesystem driver.
Right; but the maker of the OS makes the filesystem driver. To repeat myself for, I believe, the fourth time.
-
Talisker Storm
What do you think of it? I've been taking a "tour" of single malts recently. My favorite so far on the balance of smoothness and not making me bankrupt I believe is Craggonmore 12 year old.
I know you got it as a gift, but it seems weird to get a no age statement scotch from an old established distillery. I guess they're testing out a new recipe?
-
Software RAID?
Windows and OS X manage software RAID without requiring it. What the fuck's wrong with Linux?
Did you ever see that guy who made a software RAID on his OS X box out of like 250 USB 3.5" disk drives? That was awesome.
LVM?
LVM is so vague I can't really tell what it is, but it sounds like it's basically the "virtual" partition management that Windows has had since about Windows 2000. (Lets you resize partitions on the fly and such. Am I guessing right?)
That's what the Wiki page implies it is, at least. Needless to say, Windows supports that without requiring a boot partition. What the fuck's wrong with Linux?
LUKS?
Bitlocker on Windows works fine without a boot partition. What the fuck's wrong with Linux?
-
My favorite so far on the balance of smoothness and not making me bankrupt I believe is Craggonmore 12 year old.
I am big fan of Oban. Costco has it for only 48$ but IMO it is far better than overpriced Johnny Walker Blue for 160$
. I like Oban better than Lagavulin (80$) or Glenlivet (90$) but Glenlivet is also fine.
-
LVM is so vague I can't really tell what it is, but it sounds like it's basically the "virtual" partition management
It is crap Redhat wanted to be tested so made it default even though it makes absolutely no sense on laptops and most desktops. I was bitten by it once but since then learned to select ext4, I am fine testing software even from tip of git but NOT MY FILESYSTEM. They are now abandoning it for BTRFS, which is solid but I do not feel adventurous anymore :)
-
even though it makes absolutely no sense on laptops and most desktops
I use it mainly for its snapshot facility.
On my servers, I can shut down all my VMs to close their virtual disks, then take snapshots of my LVM volumes, then start the VMs up again; this takes under a minute. Then I can back up the snapshots to multiple destinations. The over-the-Internet offsite backup can take a few hours; having the source be a snapshot means that all files are in a mutually consistent state and I don't really care how long the backup takes.
In theory I could do this without even shutting down the VMs, but I like my backup snapshots to be taken in a state where all the VM disk images are internally consistent. Also, most of them are Windows VMs and therefore occasionally do weird crap of various kinds if left running for too long.
I use LVM on my home machines as well. I've found its snapshot facility useful for establishing rollback points for
apt-get dist-upgradein case of another unexpected GNOME 3 kind of disaster, and I've also used it to migrate a whole installation to new, bigger drives with no downtime, after the old drive's SMART log started showing more sector reallocations than I was happy to see.Being able to resize volumes on-the-fly is nice too, though now that terabyte-scale drives mean wasting a few tens of gig on a too-big system partition doesn't matter at all, I haven't needed to do that for a while.
It does look like btrfs is going to let me do everything I've been accustomed to achieving with ext4 on top of LVM on top of mdraid, only better and easier. I plan to wait for a couple more years until its worst-case performance is no longer unpredictably awful, and it has maintenance and repair tools as competent as ext4 fsck, then move to that instead.
A lot of what I do on Linux with LVM can also be done on Windows using NTFS shadow copies (volume snapshots). I like LVM better because I can get early warning of trouble brewing (e.g. snapshot volumes running out of disk space) and because I have never yet seen LVM do the Windows thing of discarding all snapshots of a given volume after some kind of I/O timeout event.
-
Bitlocker on Windows works fine without a boot partition. What the fuck's wrong with Linux?
Windows >= 7 on a BIOS environment will create a boot partition. It may not in UEFI mode, but as mentioned by people above linux in UEFI mode doesn't need boot either since grub-efi can live in the EFI partition.
-
Ahhh that's more like blakey . I was allready afraid after the start.
-
@asdf said:
If you have 8GB+ of RAM, you'll probably never need swap space anyway.
You don't need/need it, but you really want/need it.
Actually, you do need/need it, both on Windows and on Linux.I've tried Windows 8.1 without a page file. The result was that Windows started to kill applications for "out of memory", even though there was still 300MB of free memory and 1GB of RAM just in use for disk caches.
I've also tried Linux without a page file. Then, if the memory is full, an enormous amount of disk activity starts, sometimes locking up the whole system. Perhaps it starts to discard executable pages from memory and re-read them from the executables on disk when necessary, in order to grant requests for new data pages in memory rather than killing processes.
-
@OffByOne said:
Talisker Storm
What do you think of it? I've been taking a "tour" of single malts recently. My favorite so far on the balance of smoothness and not making me bankrupt I believe is Craggonmore 12 year old.
I like it a lot. It's smooth and has a nice aftertaste. It's somewhat peaty (but the peat is not dominating the taste, although I wouldnt mind that) with a salty accent. The "made by the sea" slogan they put on the box is not a lie.
I know you got it as a gift, but it seems weird to get a no age statement scotch from an old established distillery. I guess they're testing out a new recipe?
Funny you mention that. I thought the same thing when looking for the age statement on the box and bottle yesterday when I was writing the post you replied to.
I might be testing, it might be marketing. It tastes more mature (is that a correct description?) than some 7yo whisky I had previously.
Since it's a gift, I don't know how much it cost, especially because tax regulations on liquor are probably wildly different between the US and Belgium.
The Secret Santa group where I got the bottle has a rule that gifts should total about €30, so there is that.If you like peaty and slightly salty, give it a try.
-
It's probably the combination of legacy Unix crap and the mentality of Linux champions who are promoting these kinds of "secure" setups.
Not necessarily so. I use md-raid even on laptops to increase bandwidth (say, dual SSD setup). /boot partition is there for the bootloader and kernel, the rest is a RAID0 slice. Every single platform does it. OS X does exactly this behind the scenes when you create software RAID. Windows creates small hidden partitions for fuck knows what reasons even if you don't use advanced partitioning/logical volumes.This can be, of course, mitigated by use of EFI system partitions in newer systems, but they serve ultimately the same purpose as /boot.
-
Perhaps it starts to discard executable pages from memory and re-read them from the executables on disk when necessary, in order to grant requests for new data pages in memory rather than killing processes.
This. You can tune the OOM killer, though, to change this behavior and kill something earlier. Folks do it in Chrome OS, and I think Android does it, too.
-
LVM is so vague I can't really tell what it is, but it sounds like it's basically the "virtual" partition management that Windows has had since about Windows 2000. (Lets you resize partitions on the fly and such. Am I guessing right?)
Yeah, it's basically the same, and Linux introduced it at approximately the same time.
What the fuck's wrong with Linux?
If you just used one or two, you were fine. But stacking all three of them arbitrarily didn't work in the past. Apparently it does now. Still no big deal really, since at that time every distro installer created a separate 200MB /boot partition by default. It just annoyed me.
-
Craggonmore 12 year old
If you like Cragganmore, you'll probably like the 10-year-old Talisker as well. It's a bit more peaty, but still balanced and not as peaty as e.g. Laphroaig. So unless you absolutely hate even slightly peaty whiskys, you should try it.
-
In theory I could do this without even shutting down the VMs, but I like my backup snapshots to be taken in a state where all the VM disk images are internally consistent.
What do you mean by "internally consistent"? The whole point of LVM snapshots is that it stops writing to the partition when it's in a consistent state, let's you back up the snapshot, and then re-applies the changes made in the meantime, isn't it?
-
I might be testing, it might be marketing.
Well, for years, marketing has tried to tell us that the age statement says something about the quality of the whisky. Which is only true in some cases (<5 years tastes shitty most of the time, although I've seen exceptions to that rule)*. The only things the age statement ever told you was:
- Whether you should expect a whisky that tastes more like whisky or more like the barrel it's been in.
- How expensive producing that whisky has been for the distillery.
Also, the bottle never stated the exact age of the whisky, only the minimum age. AFAIK, the distilleries usually mix whiskys from different years to create the flavor they want.
That said, if the distilleries now stop stating the age on the bottle, they might do that for two reasons:
- Increase their profits by selling you cheap, crappy whisky (which is obviously what Macallan is doing right now).
- They want to experiment a bit by mixing "young" and "old" whisky without making everyone think it's bad stuff.
In case of Talisker, I think it's the latter, so: Both marketing (because people mistakenly think age = quality) and experimentation.
The Secret Santa group where I got the bottle has a rule that gifts should total about €30, so there is that.
For some weird reason, Talisker is way more expensive in the US than in Europe (compared to the average price for whisky on each side of the pond). Honestly, they could charge more than 30€ for it in Europe as well, since it's extremely popular and good. Not sure why they don't do that.
*) If you want an example for a whisky where the younger version is definitely better, try the 12-year-old non-filtered Aberlour and the 18 year old Aberlour.
-
LVM is so vague I can't really tell what it is, but it sounds like it's basically the "virtual" partition management that Windows has had since about Windows 2000. (Lets you resize partitions on the fly and such. Am I guessing right?)
It incorporates that capability.
LVM allows (if you want to take the time to configure it properly) you to collect physical drives and partitions into groups and then create logical volumes over those groups. In addition to allowing partition resizing, you can also configure volumes that span across physical volumes (with or without striping) and mirroring. It also allows you to quickly make read-only "snapshots" of volumes, which is useful for backups, restoring a system to a last-known-good state and other similar things.
I'm not sure about the full extent of its capabilities and I don't know how reliable it is when the more advanced capabilities are used these days (I've read about some spectacular failures when it glitches, but I think that was back when LVM was new) but it works very well for the basic capabilities that distro-installers set up for you.
As @dse wrote, it's really not necessary for people who want a basic installation on one or two partitions of a single volume, but RedHat/CentOS promotes it. Probably because they are promoting themselves as an "Enterprise" distribution, and such capabilities (assuming the implementation is solid, of course) are useful for servers.
They are now abandoning it for BTRFS
Not quite. btrfs is a file system that would be used instead of ext4. While it provides some of LVM's features (like snapshots), it is not a complete replacement. LVM would still be needed for volumes that span across drives, or for mirroring, for instance.
-
@blakeyrat said:
Craggonmore 12 year old
If you like Cragganmore, you'll probably like the 10-year-old Talisker as well. It's a bit more peaty, but still balanced and not as peaty as e.g. Laphroaig. So unless you absolutely hate even slightly peaty whiskys, you should try it.
Since we have a
thread, how about a 
Whisker thread? 99% of the whiskeys I've tasted are like paint thinner laced with AIDS needles. I'd like to find more of the 1%.
-
I've tried Windows 8.1 without a page file. The result was that Windows started to kill applications for "out of memory", even though there was still 300MB of free memory and 1GB of RAM just in use for disk caches.
Windows doesn't allow applications to commit memory over the physical amount of memory available to it. So this is "normal".
Anyway, you're just repeating what I said. So.
I've also tried Linux without a page file. Then, if the memory is full, an enormous amount of disk activity starts, sometimes locking up the whole system. Perhaps it starts to discard executable pages from memory and re-read them from the executables on disk when necessary, in order to grant requests for new data pages in memory rather than killing processes.
My understanding is Linux does allow committing more memory than is physically available to it, and when the shoe drops it just kills random processes until the commit level is back under control. Which, BTW, is excellent design. The computer equivalent of, "OH MY GOD PANIIIICICICII!!!!! FUCK USER DATA START KILLING PROCSESSES!!!"
Not sure what the disk activity would be in this case. Considering Linux doesn't "lock" open files in any way, wouldn't this:
Perhaps it starts to discard executable pages from memory and re-read them from the executables on disk when necessary,
be extraordinarily dangerous? How does Linux know the page it just loaded from disk is the same page it loaded when the application was first launched? Well since Linux is so expertly-designed, let me guess: it just assumes it's ok.
But. Again, you're just restating my point in a less pithy way.
-
Wait, 95 posts and nobody's 'shopped in a football? Jesus, this forum sucks.
-
@Lorne_Kates said:
how about a
Whisker threadNow that we have a suitable emoji, I should probably create one. I'm a lazy person, so I waited for someone else to start such a thread until now. ;)
-
Wait, 95 posts and nobody's 'shopped in a football? Jesus, this forum sucks.
Well, you're part of the people who didn't do it, so you must suck, too.
I wasn't going to try with Paint, and I don't have anything better, except paint.net, so.
-
If every coder was good at graphics there would be a lot more awesome games out there.
-
Wait, 95 posts and nobody's 'shopped in a football? Jesus, this forum sucks.
You're welcome.
