On 5/31/26 19:36, void wrote:
> On Sun, May 31, 2026 at 06:28:17PM -0700, Mark Millard wrote:
>> On 5/31/26 03:08, void wrote:
>>
>> How much swap do you use with that 320 GiBytes of RAM?
>>
>>>
>>> Here's top output during the electron41 build
>>>
>>> last pid: 76928;  load averages:   22.02,   21.41,   21.32  up
>>> 8+23:06:29  20:17:14
>>> 184 threads:   20 running, 164 sleeping
>>> CPU: 89.4% user,  0.0% nice, 10.6% system,  0.0% interrupt,  0.0% idle
>>> Mem: 84G Active, 74G Inact, 29G Wired, 1176M Buf, 188G Free
>>> ARC: 16G Total, 8693M MFU, 4833M MRU, 1368K Anon, 192M Header, 2607M
>>> Other
>>>      12G Compressed, 19G Uncompressed, 1.61:1 Ratio
>>>      Swap: 4096M Total, 4096M Free
> 
> 4096M swap

So RAM+SWAP = 324 GiBytes instead of 320 GiBytes: RAM+SWAP = 1.0125*RAM.
Seems unlikely to make much of a difference in the failure rate
associated with running out of swap space, such as via tmpfs competition
for RAM+SWAP.

For 64-bit contexts, I use more like RAM+SWAP = 4.75*RAM when I have the
overall storage to not force a tradeoff. The one place where a tradeoff
was force I have more like RAM+SWAP = 3.66*RAM. I use vm.pageout_oom_seq
to generally avoid OOM kills before running out of RAM+SWAP. 4.75 or so
(or less) avoids FreeBSD warning of potential mistuning for my RAM
contexts from 8 GiByte to 192 GiByte. As I remember, sufficiently
smaller RAM sizes needed more like 4.6 or 4.5 to avoid the warning. I've
probably not ever used less than 1 GiByte system for such activities.

Because of my very rare `bulk -ca` experiments, I blacklist the tmpfs
use of what would otherwise be large usage of tmpfs, as well as doing
some mutual exclusion. Otherwise I build less of a variety of such than
you do.

I also avoid spinning rust storage media.

-- 
===
Mark Millard
marklmi at yahoo.com

Reply via email to