Practicality of older machine running Gentoo.

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

I've had problems with portage and autofs not automounting, but I think that has something to do with portage's namespace sandboxing. If I work around that (mount it before running emerge) it hasn't caused any other problems.

eccerr0r · Posted: Tue Dec 10, 2019 11:54 pm Post subject:

Every time I try building on NFS, portage constantly gives me files it can't delete for some reason. Never had an issue on a filesystem on a block device...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

That sounds like a good reason to avoid it, yeah... I'm using NFSv4.1-only, don't know if that makes a difference.

technotorpedo · Apprentice Joined: 10 Dec 2019 Posts: 151

eccerr0r · Posted: Wed Dec 11, 2019 7:47 am Post subject:

The NFS problems are race condition problems and not permission problems. Not only that, it still sort of works but gives a lot of diagnostic output indicating potential issues though a lot can be ignored and cleaned up later. Pretty much the only program that has issues with nfs is portage, I've not seen anything else fail on nfs, my home directory is on nfs as well, and even sendmail/procmail locking works fine.

Now I don't know about NFSv4 behavior, it may be better. I've been using NFSv2/3 since forever and haven't gotten around to upgrading.

Kind of strange, seems people frequently put /usr/portage and especially /usr/portage/distfiles on nfs but not PORTAGE_TMPDIR ...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

erm67 · l33t Joined: 01 Nov 2005 Posts: 653 Location: EU

The problem with compiling the kernel, or whatever uses make is that nfs makes heavy use of caching async i/o, in particular usually the problem is that:
1) the timestamp of files is not guaranteed on nfs, the server might return an approximation of it to the client before the files is actually written on disk even if the clocks of server and client are in ntp-sync
2) the client might cache the wrong timestamp until the file attributes cache expires,
3) the client updates the file attributes from the server, that in the mean time has stored the file on the backing file system and returns the correct and possibly different timestamp
4) make gets all upset because the timestamp of a file has changed

That is just how nfs work, it might get better if you mount using sync and noac but it gets so horribly slow that is unusable. or maybe try to increase cache timeout on the client (actimeo) to a very long time so that the timestamps are not re-read from the server during compilation.
_________________
Ok boomer
True ignorance is not the absence of knowledge, but the refusal to acquire it.
Ab esse ad posse valet, a posse ad esse non valet consequentia

My fediverse account: @erm67@erm67.dynu.net

eccerr0r · Posted: Wed Dec 11, 2019 3:40 pm Post subject:

Yeah, I'm pretty sure mounting it sync would fix things, even v2/v3, but it'd slow things down significantly.
Not sure how other people are using nfs for portage tmpdir, is this how people do this? I can't see how this can only be affecting one person.

Also why is portage depending on timestamps on the scratch tmpdir, perhaps portage should use two tmpdirs - one for locks on the local filesystem that guarantees operations being atomic and monotonic, and all other noncritical "bulk" stuff goes on the other.

Bother bother. After 17 hours for rust (not including the failed runs), it's building firefox now. At least firefox can partially be distcc'ed but the load average is too high due to rust taking more resources than make assumes...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

erm67 · l33t Joined: 01 Nov 2005 Posts: 653 Location: EU

With nfs4 you can get rid of most of the portmap/rpcbind non-sense and save some ram :-)

unfortunately mountd is still required even if nobody really understands why .... Maybe nfs5 will be fully free of that ancient stuff.
_________________
Ok boomer
True ignorance is not the absence of knowledge, but the refusal to acquire it.
Ab esse ad posse valet, a posse ad esse non valet consequentia

My fediverse account: @erm67@erm67.dynu.net

eccerr0r · Posted: Wed Dec 11, 2019 7:42 pm Post subject:

Fortunately RAM isn't that bad with 2GiB for a single core, just need to be careful and not make it thrash when make/ninja gets confused and runs too many big processes.

Anyway, the for the Atom's update is done for now including kernel 4.19.86 which I distcc'ed.

Until next time... or if I feel like updating my 1GiB Pentium-M 1.6GHz next. It's unfortunately kind of flaky with 1.5GiB RAM...

BTW, I thought that portmap/rpcbind though annoying to configure, was really a system's DPNS or dynamic port name server (contrast and similarities to the much more common dynamic DNS). Helps to prevent collisions between different services trying to get the same port. Other than configuration, it's probably not that big of a deal to have around, though nowadays having a fixed port is more common though ensuring no other application uses it can be the challenge (hey, why can't I use port 80 for my whizbangsoftware?)
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

erm67 · l33t Joined: 01 Nov 2005 Posts: 653 Location: EU

Remote procedure call
everybody dreams that other people runs programs remotely on their server :-)

Using only port 2049 for everything also eases firewall rules.
NFS4 doesn't require it but some ancillary protocols like the nfs mount protocol were not updated.

recently a change in nfs-utils made rpc.mountd require 32Mb resident memory which is really bad if you only have 128Mb on the NAS and it is used only to mount the share, I used samba for a while but in the end updated the NAS. Samba with posix extensions can be faster than nfs and less resource hungry, that really say everything about the current state of NFS. Samba also does everything using a couple of ports.
_________________
Ok boomer
True ignorance is not the absence of knowledge, but the refusal to acquire it.
Ab esse ad posse valet, a posse ad esse non valet consequentia

My fediverse account: @erm67@erm67.dynu.net

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

erm67 · l33t Joined: 01 Nov 2005 Posts: 653 Location: EU

figueroa · Posted: Thu Dec 12, 2019 6:41 pm Post subject:

I have a single core Celeron with 4 G RAM (the extra memory really helps and can often be found really inexpensive these days) which I update only every couple of months. XFCE4 desktop, Firefox-bin Thunderbird-bin, LibreOffice-bin and otherwise little fluff. Updates usually run for a couple of days and I don't usually have to attend to them. These days, Portage is really smart. I use thsi small laptop mainly when I travel and it's a pleasure to use.
_________________
Andy Figueroa
hp pavilion hpe h8-1260t/2AB5; spinning rust x3
i7-2600 @ 3.40GHz; 16 gb; Radeon HD 7570
amd64/23.0/split-usr/desktop (stable), OpenRC, -systemd -pulseaudio -uefi

eccerr0r · Posted: Thu Dec 12, 2019 7:01 pm Post subject:

Which "Celeron" is it? clock speed and architecture?
Some Celerons are fairly fast, but pretty much all Atoms are universally slow...

Heck the Celeron-M 1.5GHz handily beats my Atom 1.6GHz (both have 2GiB RAM, though the Atom's SSD makes disk io comparisons unfair), and there are other CPUs branded as Celeron that are even faster.

That being said, there are Celerons that use Atom cores...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

figueroa · Posted: Thu Dec 12, 2019 7:51 pm Post subject:

eccerr0r · Posted: Thu Dec 12, 2019 8:18 pm Post subject:

Based on architectural improvements of Core2 over Pentium-M, I'd say the SU3500 is still faster than the Celeron-M 1.5GHz, which is faster than the Atom 1.6GHz. But yeah agreed it's not that much faster.

That 4GiB RAM I think is minimum if you're building 64-bit. I've pretty much given up on 2GiB 64-bit, it swaps badly during build even using 1 core where 32-bit is fine.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

technotorpedo · Apprentice Joined: 10 Dec 2019 Posts: 151

Thanks for the interesting info guys, not at all a nfs expert. Outside of generally knowing what the hades it even is, shrugs. Though there ya go OP, network boot, boot server etc ? Have zero idea if that's practical or useful to you for what you're dealing with. Imagine something like Gentoo can be custom tailored to just about any niche or use, to a point of being better, at least surely as good as anything else gnu/Linux. On a given isolated system not sure but being gnu/Linux there's almost NEVER a shortage of options. Just a matter of settling on what works best for x-user/usecase. According to needs + preferences.

e3k · Guru Joined: 01 Oct 2007 Posts: 513 Location: Inner Space

got a "new" machine. intel core i5 2520m. compiling at snail speed after selecting the desktop profile. got to get distcc running. probably i should have done this before selecting the desktop profile? because now when i emerge distcc it pulls 40 packages...
...
first i thought it will be easy to setup distcc. then i thought impossible. now it is somehow (compile error with g++ versions mismatch 9.2.0 vs. 8.3.0) running debian <-> gentoo. lets see if firefox executes after it is done.
...
that rustc is an overkill for the poor old sandybridge. cpu 100% and half of RAM eaten up. very very slow. the compilation of the dependencies took an afternoon today *sigh*
...
i give up. distcc does not do rust or vice versa. also gcc and g++ not distributed. stopping distccd on both machines and wifi on the gentoo host.
is there any non rust based browser? i know only links...
...
[OK] the compilation of the package firefox did stop after almost 2h. does it mean the broken distcc slows things down or am i just expecting too much from a 10y old CPU? Next time i use binary as last resort.
...
updating debian to testing provided me the current version of gcc. now distcc does also g++ and gcc. also tuning the -j and -l flags helped me to use all the cores. i had to add -fPIC compile flag on the sandy bridge to be able to successfully compile.
_________________

Flux & Contemplation - Portrait of an Artist in Isolation

eccerr0r · Posted: Tue Mar 02, 2021 7:34 am Post subject:

Sorry for being >1 year late, but:

gcc, g++, and clang(!!!) should work over distcc along with their associated cross compilers. One thing that may be useful to run "distcc-config --update-masquerade" to update the fake gcc directories. Also note you should uninstall old versions of clang/llvm prior to 10, it seems to really mess up distcc due to how update-env is handled for old versions of clang/llvm.

But you're right, distcc does not distribute rust commands. You need distrust which still doesn't exist yet even now, as far as I know.
My quad core sandybridge completes rust in under 2 hours and yes it uses a lot of RAM. Currently it's less than 1h 30min as I distcc (to a few core2 cores) the clang/llvm portions of rust, but the other portions of rust cannot be distributed. Firefox completes in just over the hour mark (again with distcc - which again should also help the llvm/clang portions). I'd expect a dual core sandybridge to take twice as long as my quad core sandbridge, and maybe more as I have some multicore distcc helpers.

I think chromium is still all gcc or clang.

And be patient with your builds. My atom-1.6GHz takes about a day to finish compiling rust even with helpers for the llvm/clang portions...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

e3k · Guru Joined: 01 Oct 2007 Posts: 513 Location: Inner Space

thanks eccerr0r but i gave up. on this slow machine i am running now ARCH.
_________________

Flux & Contemplation - Portrait of an Artist in Isolation

Fitzcarraldo · Posted: Tue Mar 02, 2021 3:08 pm Post subject:

Until end 2013 I ran Gentoo with Xfce on an old Gateway Solo 9300 laptop:

- Intel Mobile Pentium III Coppermine (800 MHz)
– 288 MB DRAM
– ATI RAGE Mobility-P, 16 MB SGRAM
– 10 GB HDD
– Belkin F5D5010 CardBus Network Card (wired Ethernet)
– Linksys WPC54G (EU) v7.1 wireless notebook adapter

I installed Gentoo just for the fun of it, and it worked quite well for a few years but upgrading gradually became more and more time-consuming and 'painful'. I just wouldn't bother with Gentoo on something like that now.

I still have my ASRock ION 330HT nettop, which used to be my HTPC until smart TVs made it obsolete:

– Intel Atom 330 (1.6 GHz)
– 4 GB DDR2 (the maximum the machine can take)
– NVIDIA ION Graphics
– 320 GB HDD
– NVIDIA MCP79 Gigabit Ethernet controller
– Atheros Communications AR9285 Wireless Network Adapter

With two 2GB RAM modules fitted the nettop detects 3327MB of RAM. From what I can glean on the Web, this is a limitation of the motherboard or BIOS in this nettop, because the Intel Atom 330 CPU itself has a 64-bit instruction set and is capable of supporting 4GB of RAM. Anyway, that machine has Lubuntu installed on it at the moment, and I wouldn't dream of installing Gentoo on that machine either; in my opinion it's a waste of time and energy compiling packages on something like that when lightweight binary distributions can be installed and updated easily and quickly these days, and work very well indeed (at least on the three machines of mine that have binary distributions installed). I keep Gentoo on my two laptops with Core i7 CPUs and plenty of RAM. Actually, my old Compal NBLB2 laptop (with max permissible 8 GB RAM and first-generation Core i7 CPU) struggles with Gentoo and KDE upgrades these days, and I have been tempted to install a binary distribution on it instead, but have resisted so far as I am a sucker for punishment (and I like Gentoo)!
_________________
Clevo W230SS: amd64, VIDEO_CARDS="intel modesetting nvidia".
Compal NBLB2: ~amd64, xf86-video-ati. Dual boot Win 7 Pro 64-bit.
OpenRC udev elogind & KDE on both.
Fitzcarraldo's blog

eccerr0r · Posted: Tue Mar 02, 2021 4:23 pm Post subject:

I have a hard time believing even modern machines are compiling rust in much less than 1 hour(unless you have something like a 6-core or more? Clock speeds and IPC don't seem to be getting that much faster anymore.)

This machine gets distcc and nbd (for swap)/nfs(for disk) help but yes it's still being updated/used (I'm posting this message with this machine!):