[Nix-dev] i686 Builds?

Christian Theune ct at flyingcircus.io
Tue May 12 12:05:29 CEST 2015


Hi,

same here.

Many interpreted languages (like Python) are affected by this as they tend to be quite pointer-happy. As pointer-size doubles from 32bit to 64bit we find that in most applications we have about 70% increase when moving to 64-bit ending up with 1.7 as much memory as before. So we also currently run applications in 32-bit virtual machines and rather use many 3GiB processes than a few bigger ones. Moving from 3GiB to 64bit requires about 5GiB just to even out the pointer-size effects.

Supposedly the amd64 instruction set has some benefits that make e.g. Python run faster on certain computational stuff, but I don’t have prove for that.

In the long term we will include 64-bit in the mix anyway as some applications (Mongo, sigh) are quite trigger happy with allocating virtual (non residential) memory for mmapping insane numbers of insanely large files …

Christian

> On 12 May 2015, at 11:59, Lluís Batlle i Rossell <viric at viric.name> wrote:
> 
> My experience is equal with Marco, about memory and my usage of i686. i686
> is important for me too.
> 
> On Tue, May 12, 2015 at 11:43:47AM +0200, Marco Maggesi wrote:
>> I use 32 bit a lot.
>> First of all, I use it on some old machines with 32bit hardware.
>> But, more importantly, I use it regularly on virtuabox and xen virtual
>> machines.
>> In my experience, for most of my use cases the 32bit require less memory
>> (which is often not abundant on virtual instances) and it is thus generally
>> faster for many computing tasks .  I made some tests with HOL Light (the
>> theorem prover).  The bare program has memory occupation which almost the
>> double in the 64bit version (~1.2Gb) with respect to the 32bit version
>> (~0.7Gb).  On a virtual machine with 2Gb of ram, the 32 bit it is often
>> 10%-20% faster on typical usage and 50% faster or more when the computation
>> requires more memory.
>> In my experience, the version 32 bit can be more convenient than the 64bit
>> version in a variety of situations.
>> So, please, do not give-up with 32 bit support.
>> Marco
>> 
>> 
>> 
>> 2015-05-12 11:08 GMT+02:00 Luke Clifton <ltclifton at gmail.com>:
>> 
>>> +1
>>> 
>>> This seems like a good idea.
>>> 
>>> On 12 May 2015 at 06:45, William Kennington <william at wkennington.com>
>>> wrote:
>>> 
>>>> Maybe it would make more sense to only build the i686 builds if our
>>>> tested set of x86_64 binaries build correctly. We would still release with
>>>> both but it would cut down on a lot of redundant failures.
>>>> 
>>>> On Mon, May 11, 2015 at 3:39 PM Ryan Trinkle <ryan.trinkle at gmail.com>
>>>> wrote:
>>>> 
>>>>> I encountered an i686 user just the other day!  I don't use it
>>>>> personally, but having solid support in Nix was fantastic, especially
>>>>> because older, 32-bit machines tend to be slower, which makes Nix's binary
>>>>> caching functionality even more important.
>>>>> 
>>>>> On Mon, May 11, 2015 at 6:36 PM, Shea Levy <shea at shealevy.com> wrote:
>>>>> 
>>>>>> Hi all,
>>>>>> 
>>>>>> Do we still have users running 32-bit machines? It would reduce the
>>>>>> load on
>>>>>> hydra significantly if we could drop support for i686, though of course
>>>>>> if
>>>>>> people are still relying on it we shouldn't make the change yet.
>>>>>> 
>>>>>> ~Shea
>>>>>> _______________________________________________
>>>>>> nix-dev mailing list
>>>>>> nix-dev at lists.science.uu.nl
>>>>>> http://lists.science.uu.nl/mailman/listinfo/nix-dev
>>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> nix-dev mailing list
>>>>> nix-dev at lists.science.uu.nl
>>>>> http://lists.science.uu.nl/mailman/listinfo/nix-dev
>>>>> 
>>>> 
>>>> _______________________________________________
>>>> nix-dev mailing list
>>>> nix-dev at lists.science.uu.nl
>>>> http://lists.science.uu.nl/mailman/listinfo/nix-dev
>>>> 
>>>> 
>>> 
>>> _______________________________________________
>>> nix-dev mailing list
>>> nix-dev at lists.science.uu.nl
>>> http://lists.science.uu.nl/mailman/listinfo/nix-dev
>>> 
>>> 
> 
>> _______________________________________________
>> nix-dev mailing list
>> nix-dev at lists.science.uu.nl
>> http://lists.science.uu.nl/mailman/listinfo/nix-dev
> 
> 
> --
> (Escriu-me xifrat si saps PGP / Write ciphered if you know PGP)
> PGP key D4831A8A - https://emailselfdefense.fsf.org/
> _______________________________________________
> nix-dev mailing list
> nix-dev at lists.science.uu.nl
> http://lists.science.uu.nl/mailman/listinfo/nix-dev

—
Christian Theune · ct at flyingcircus.io · +49 345 219401 0
Flying Circus Internet Operations GmbH · http://flyingcircus.io
Forsterstraße 29 · 06112 Halle (Saale) · Deutschland
HR Stendal HRB 21169 · Geschäftsführer: Christian. Theune, Christian. Zagrodnick

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 496 bytes
Desc: Message signed with OpenPGP using GPGMail
Url : http://lists.science.uu.nl/pipermail/nix-dev/attachments/20150512/e395db8b/attachment.bin 


More information about the nix-dev mailing list