Dropping i686 support

masteryod · 2012-09-05 11:00:47

What a coincidence amd64/x86_64 just became most popular architecture in Debian.

Source and some interesting statistics: http://www.thepowerbase.com/2012/09/amd … hitecture/

stratabum · 2012-09-05 20:16:38

just my experience:
my laptop was bought on a very tight budget a bit more than 2 years ago, it came with a 64bit capable cpu (almost a celeron, *grin*) and just 2gb of ram (oem installed freedos, cheap cheap I know), basically a low end laptop dressed in a netbook price (disclaimer: I believe the rest of this post could be applicable to this same latter numerous category too).

cheap it was cheap, but I shopped it wisely as the screen is a very relaxing matte one, it has good ergonomics, integrated bluetooth, even a coffee spillable keyboard: it has all I need and no more so buying a newer machine it's not an option, yet (in the eu country where I live unemployement rate it's still a bit up but that's another story...).

after initially installing another 64bit flavored mainstream distro and having hit some walls using wine, virtualbox, flash, integrated intel graphics, wireless drivers, some proprietary binary blobs, few other issues and seeing the already scarce ram being eaten for nothing (as 64bit oss are memory hungrier), I switched to the good ol' arch, using the 32bit version this time after trying the 32bit one of the same other distro and having finally seen some light...

add, maybe, a poor initial partitioning scheme, tune this and that, everything's now working (since the dawn of linux, not all the laptops could say the same), it's perfectly tailored and scripted to my very needs, works like a charm, stays up for months (as unix is meant to be) and all the rest.

I hope it will survive quite few more years.

today I received a 4gb ddr3 dimm, a very cheap one of course: shutdown, unscrew, insert, screw back, bios confirmation of 6gb and boot.

do you (really) think I should now ditch my beloved os, backing up (hopefully) everything (even all the bits sprinkled here and there), reinstall (64bit this time, lots of aur and some other painfully compiled stuff) and reconfigure everything because apache or gcrypt will be way faster and praying for the best?

well, maybe under threat!

*hint*:
$ yaourt -S linux-pae

also, give /tmp at least 2gb of space (see here for more infos) as 1,5gb (default half recognized ram not yet pae enabled) wouldn't suffice me to recompile today's kernel (3.5.3-1).

and don't forget to add a stanza (i.e. an entry in ibm parlance) to the /boot/grub/menu.lst for the new kernel and initramfs or any other bootloader config file you're using.

cheers,
nix

ZekeSulastin · 2012-09-06 18:50:14

There are not big enough desks in the world for me to express the necessary headdesking at this post, stratabaum. Way to ignore pretty much everything ever about x86_64 and PAE.

stratabum · 2012-09-08 12:44:29

ZekeSOulastin, if you meant my post was too long, here is the summary:

- have you already installed and configured Arch 32bit on a PC and are happy with it?
- have you just upgraded over 3Gb of RAM?
- you're not forced to reinstall from scratch the 64bit version, the hassle could not be worth.
- install instead linux-pae and live merry!

cheers,
nix

usagi · 2012-09-10 02:04:17

Stratabaum: Your post was too long, but i thinkZekeSulastin was 'headdesking' about actually recommending using PAE as an alternative to 64-bit systems. PAE was and is a bad hack to cheat around addressing physical memory. It adds overhead and gains you nothing but headache and eventual major problems a process ever uses memory in a specific, unusual manner.

To put it colloquially:

Using PAE on an i686 Linux to avoid a reinstall is like duct taping wings to an old Pinto because you think airport security is inconveienient.

mrechte · 2012-09-10 12:22:37

usagi wrote:

Using PAE on an i686 Linux to avoid a reinstall is like duct taping wings to an old Pinto because you think airport security is inconveienient.

I have been using Linux PAE for a few years on a machine which has many services running, including several VM (Win XP, RHEL) and I am very happy with it. The only problem is that it does not come pre-compiled on arch, but I got used to build it from aur (http://aur.archlinux.org/packages.php?ID=51326). On the other hand on a 64 bit Linux I got troubles with some software (Unix ODBC drivers) coming in two flavours (32 bits and 64 bits) which got confused when loading their libraries. Also considering that none of my services require more than 2 GB individually, I understand there is a quite lot of memory waste having 64-bit memory pointers everywhere.

Allan · 2012-09-10 12:34:12

If you have x86_64 hardware and are running i686, just install the x86_64 kernel and add "init=/usr/bin/linux32 /sbin/init" to your kernel line in your bootload configuration. That is a "poor man's" PAE...

Leonid.I · 2012-09-10 15:10:42

mrechte wrote:

usagi wrote:
Using PAE on an i686 Linux to avoid a reinstall is like duct taping wings to an old Pinto because you think airport security is inconveienient.
I have been using Linux PAE for a few years on a machine which has many services running, including several VM (Win XP, RHEL) and I am very happy with it. The only problem is that it does not come pre-compiled on arch, but I got used to build it from aur (http://aur.archlinux.org/packages.php?ID=51326). On the other hand on a 64 bit Linux I got troubles with some software (Unix ODBC drivers) coming in two flavours (32 bits and 64 bits) which got confused when loading their libraries. Also considering that none of my services require more than 2 GB individually, I understand there is a quite lot of memory waste having 64-bit memory pointers everywhere.

You assume that PAE implementation in the kernel is at the same level and of the same quality as x86_64. In fact, it's not. See http://cl4ssic4l.wordpress.com/2011/05/ … -about-pae

wombat23 · 2012-09-28 11:01:35

blackout23 wrote:

I think the first consumer CPUs that were 64 bit capable have been released in 2002. So unless your PC is +10 years old you should use x86_64.

The first AMD consumer CPU with 64 bit support was the Athlon 64, released in late 2003.
Intel's first 64 bit consumer CPU was an updated Pentium 4 in 2004.

Even the first "Intel Core" series CPUs sold in 2006 had no 64 bit support at all.

So it's rather +6 years, not +10.

D4ve · 2012-09-28 20:59:13

I'm not sure, but aren't there Intel Atom processors without 64bit support?

That could be a problem too

Last edited by D4ve (2012-09-28 20:59:28)

Gusar · 2012-09-28 21:05:54

D4ve wrote:

I'm not sure, but aren't there Intel Atom processors without 64bit support?

The very first Atoms (Diamondville) are 32bit, yes. I'm typing this on one . Starting with Pineview, the Atoms are 64bit. But that doesn't matter much, there are much older machines than Diamondville still in regular use. For example there's plenty of Pentium-M and P4 machines still around.

cgarcia · 2012-10-01 19:30:17

The question is... will it be so huge boost in multimedia applications for x86_64, to justify dropping the compatibility with many netbooks sold from 2009? (like my own eee PC 1000H for example). As others have commented, even laptops with core 2 duo are not 64-bit enabled.

I assume that having separate compilation settings for 64/32 bits would be a more heavy packaging/testing work. Perhaps we should choose which multimedia applications will really get huge benefits, and drop the 32 bit support only for them. However, this may not be good for all (I use my netbook as wifi AP, but others may use multimedia with them).

Leonid.I · 2012-10-02 21:57:44

cgarcia wrote:

The question is... will it be so huge boost in multimedia applications for x86_64, to justify dropping the compatibility with many netbooks sold from 2009? (like my own eee PC 1000H for example). As others have commented, even laptops with core 2 duo are not 64-bit enabled.

This is not about performance. In fact, 64bit is only beneficial performance-wise if the application is specifically using 64bit registers. The advantage comes in the form of stability (due to larger address space) and security (due to NX bit).

Basically, the rule-is simple: "if CPU supports 64bit and RAM >= 1GiB, use x86_64 OS". I know things like skype, flash player or acrobat reader do not run natively on x86_64. But this means these are jusp bad programs -- no need to sacrifice the OS.

Also, if a machine with C2D can't run in 64bit mode, there is a bug in BIOS which should be addressed. Dropping down to i686 is not a solution.

levir3 · 2013-02-03 15:40:52

I' ve decided to install linux on our 8 years old dell (x86_32) latitude. We still keep this thing next to our Apples for all sorts of things.
It still works properly but a fast and simple os will suit this machine better then Windows. Therefore Arch is on top of my shortlist. (Debian is second on quite a distance.)
I thus hope to keep this baby alive at least for another few years.
However, this whole discussion brings me to the conclusion that i686 support will be dropped some day in the future for all the reasons mentioned.
My question is: when will Arch do this?

I'm not asking for guarantees, but to avoid unwanted distrohopping in the foreseeable future I would appreciate some sort of indication from the core developpers team whether or not i686 will be supported for at least 3 years from now.
I guess many other owners of really old harware too.

Pierre · 2013-02-03 16:15:40

i686 has no future. ATM only 8% of our users have 32bit hardware. No decisions have been made, but I would assume that in a few years this number could drop below a point where it's worth to support. We are already at the point where i686 is less tested.

graysky · 2013-02-03 16:28:23

@Pierre - Can you explain the difference between:

1) "Submissions per architectures" of which 23 % is i686.
2) "Submissions per CPU architectures" of which 8 % is i686.

https://www.archlinux.de/?page=PackageStatistics

Last edited by graysky (2013-02-03 16:28:48)

Pierre · 2013-02-03 16:30:45

1) Architecture of the operating system
2) The actual architecture the CPU supports.

So quite a few run 32bit Arch on 64bit hardware.

ConnorBehan · 2013-02-04 04:07:50

Which makes sense because a lot of people use binary blobs that would depend on all the lib32 stuff if they used 64 bit Arch. @levir3: My guess (based on how friggin easy it is to support i686 if you already support x86_64) is that i686 support will go on for another 10 years.

Allan · 2013-02-04 04:39:07

ConnorBehan wrote:

@levir3: My guess (based on how friggin easy it is to support i686 if you already support x86_64) is that i686 support will go on for another 10 years.

I'd be surprised if it was that long. In fact, I'd be very surprised if i686 support remained at the same level as x86_64 (like currently in Arch) over the next couple of years... It is already far less tested by the devs, and [core] packages rarely get more than one signoff for i686.

Edit: that is not to say it would be completely dropped... there is lots of in between ground.

ConnorBehan · 2013-02-04 04:45:47

No testing wouldn't bother me because most packages will work on i686 if they work on x86_64. I haven't tried this but it seems like one could almost maintain an i686 repo with an automated script. It would copy Arch's existing x86_64 packages and leave an MOTD warning me about certain ones that had $CARCH specific hacks. I will try this if it does get completely dropped.

Allan · 2013-02-04 04:49:49

I said the same thing... and the next day I got hit with two i686 specific bugs. So they do exist

olive · 2013-02-04 09:09:20

Just to say that I am using arch on a old i686 machine (with 2Gb of RAM) (old Celeron M 1.6 Ghz). This works perfectly fine. Recent Gnome3 is a little slow (although it works) but Xfce works like a charm. Since I don't like Gnome3 anyway, I do not see why I should upgrade to run something that I don't like. I am sure it is possible to run Arch on older machines. This machine run Firefox Libreoffice and all modern softwares without any problems of performances. I am glad to read that Arch will continue to support i686 machines and hope this is true. Moreover I think that 32 bits processor are still released now (in some notebooks). I think that supporting i686 at least for 5-10 years after the latest i686 processor was manufactured is a good idea.

A minority of people seem to think that dropping support for i686 is "bleeding edge". I would suggest they use a 128 bits only distribution, then. Surely this will be "bleeding edge"

Last edited by olive (2013-02-04 09:22:42)

olive · 2013-02-04 09:18:57

Allan wrote:

I said the same thing... and the next day I got hit with two i686 specific bugs. So they do exist

But was an Archlinux bug or an upstream bug? If it come a time where a lot of upsteeam packages begin not supporting i686 anymore, then Arch would somewhat be forced to drop i686 support; but I don't think it will happen anytime soon.

Allan · 2013-02-04 09:58:41

No - it was an upstream bug. And it is not upstream support that is lacking. It is the fact that almost no devs run i686, so no-one is testing the packages. This will get worse...

olive · 2013-02-04 10:03:59

Allan wrote:

No - it was an upstream bug. And it is not upstream support that is lacking. It is the fact that almost no devs run i686, so no-one is testing the packages. This will get worse...

But don't they run i686 at all? Even not on a chroot for testing purpose? It's not because you have encountered one i686 bug that we have to conclude that all upstream developers don't care about i868 anymore.

Last edited by olive (2013-02-04 10:04:15)

Arch Linux

#51 2012-09-05 11:00:47

Re: Dropping i686 support

#52 2012-09-05 20:16:38

Re: Dropping i686 support

#53 2012-09-06 18:50:14

Re: Dropping i686 support

#54 2012-09-08 12:44:29

Re: Dropping i686 support

#55 2012-09-10 02:04:17

Re: Dropping i686 support

#56 2012-09-10 12:22:37

Re: Dropping i686 support

#57 2012-09-10 12:34:12

Re: Dropping i686 support

#58 2012-09-10 15:10:42

Re: Dropping i686 support

#59 2012-09-28 11:01:35

Re: Dropping i686 support

#60 2012-09-28 20:59:13

Re: Dropping i686 support

#61 2012-09-28 21:05:54

Re: Dropping i686 support

#62 2012-10-01 19:30:17

Re: Dropping i686 support

#63 2012-10-02 21:57:44

Re: Dropping i686 support

#64 2013-02-03 15:40:52

Re: Dropping i686 support

#65 2013-02-03 16:15:40

Re: Dropping i686 support

#66 2013-02-03 16:28:23

Re: Dropping i686 support

#67 2013-02-03 16:30:45

Re: Dropping i686 support

#68 2013-02-04 04:07:50

Re: Dropping i686 support

#69 2013-02-04 04:39:07

Re: Dropping i686 support

#70 2013-02-04 04:45:47

Re: Dropping i686 support

#71 2013-02-04 04:49:49

Re: Dropping i686 support

#72 2013-02-04 09:09:20

Re: Dropping i686 support

#73 2013-02-04 09:18:57

Re: Dropping i686 support

#74 2013-02-04 09:58:41

Re: Dropping i686 support

#75 2013-02-04 10:03:59

Re: Dropping i686 support

Board footer