| 1 |
On Sat, Sep 1, 2012 at 8:20 PM, Brian Harring <ferringb@×××××.com> wrote: |
| 2 |
> On Fri, Aug 31, 2012 at 11:12:44AM -0400, Alexis Ballier wrote: |
| 3 |
>> On Fri, 31 Aug 2012 15:45:21 +0100 |
| 4 |
>> Ciaran McCreesh <ciaran.mccreesh@××××××××××.com> wrote: |
| 5 |
>> |
| 6 |
>> > On Fri, 31 Aug 2012 10:21:15 +0200 |
| 7 |
>> > Ulrich Mueller <ulm@g.o> wrote: |
| 8 |
>> > > Coming back to this old topic [1]. Is there still consensus that we |
| 9 |
>> > > should have such an EJOBS variable? (It shouldn't be called JOBS |
| 10 |
>> > > because this name is too generic, see the old discussion.) Then we |
| 11 |
>> > > could add it to EAPI 5. |
| 12 |
>> > > |
| 13 |
>> > > Ulrich |
| 14 |
>> > > |
| 15 |
>> > > [1] |
| 16 |
>> > > <http://archives.gentoo.org/gentoo-dev/msg_750e33f68b16d971dff1f40dd9145e56.xml> |
| 17 |
>> > |
| 18 |
>> > If we're doing this, do we tell users to stop setting MAKEOPTS for |
| 19 |
>> > EAPIs 5 and greater? |
| 20 |
>> |
| 21 |
>> How can this work ? I cant think of any simple solution. |
| 22 |
>> |
| 23 |
>> > Do we change the name of MAKEOPTS for EAPIs 5 and |
| 24 |
>> > greater instead? Do we put fancy code in the package mangler to deal |
| 25 |
>> > with it? |
| 26 |
>> |
| 27 |
>> IMHO EAPI-5 compliant PMs should do MAKEOPTS="$MAKEOPTS -j$EJOBS" for |
| 28 |
>> every EAPI; using EJOBS from ebuilds/eclasses is allowed only in EAPI 5 |
| 29 |
>> and greater. |
| 30 |
>> This is retroactive but could be classified 'PM internals' so its fine |
| 31 |
>> imho. |
| 32 |
> |
| 33 |
> This approach is fine imo, although I'd *potentially* look at adding a |
| 34 |
> magic $PROC_COUNT var that is the # of cpu threads on the system; |
| 35 |
> either that or defaulting jobs to it. |
| 36 |
> |
| 37 |
> I rather dislike requiring users to go jam a 2/4/8 in there when it's |
| 38 |
> easy to compute. That said, it's minor. |
| 39 |
|
| 40 |
On this point, I use dynamic load-based job counting. I'd be using |
| 41 |
distcc, too, but I don't have the spare hardware for a second build |
| 42 |
box right now. |
| 43 |
|
| 44 |
For a single box, I target a load average of $PROC_COUNT, with a max |
| 45 |
jobs in the vicinity of 2x$PROC_COUNT, to prevent I/O hangs from |
| 46 |
causing job count explosions. (Where I/O hiccups aren't a concern, |
| 47 |
I've found leaving --jobs open-ended worked fine, but builds always |
| 48 |
eventually expand to consume available I/O as upstream things get |
| 49 |
larger) I find this works very well; emerge and make both manage to |
| 50 |
lurk near the optimum point of keeping the CPU busy without spending |
| 51 |
too much time in context switches. |
| 52 |
|
| 53 |
In a distcc context, I would most likely target a load average of |
| 54 |
$LOCAL_PROC_COUNT, with a max jobs of |
| 55 |
(2*$LOCAL_PROC_COUNT+$REMOTE_PROC_COUNT). |
| 56 |
|
| 57 |
If ebuilds are try to be more aware of parallel-build contexts, I |
| 58 |
think it's important they're not limited to static job counts. |
| 59 |
|
| 60 |
-- |
| 61 |
:wq |
Archives