RPI3 swap experiments

[Mark Millard]

On 2018-Jul-18, at 12:09 PM, bob prohaska <fbsd at www.zefox.net> wrote:

> On Wed, Jul 18, 2018 at 07:42:13AM -0700, Mark Millard wrote:
>> 
>> 
>> On 2018-Jul-17, at 11:06 PM, bob prohaska <fbsd at www.zefox.net> wrote:
>> 
>>> It appears that some progress has been made in getting swap working reasonably
>>> on the RPI3. A -j4 buildworld attempt running r336356 to compile 336431 failed
>>> with "out of swap" but the worst read and write delays were less than 5 seconds,
>>> a marked improvement over previous examples.
>> 
>> Attributing the time variations that have been observed mostly to FreeBSD and not
>> mostly to the device at issue seems to have little or no evidence to support it.
>> 
> Possibly a fair objection. This test is with a USB3.0 flash drive. I'll repeat soon
> as possible with a USB3.1 device, which in the past reported much greater (15 second)
> delays.
> 
>>> In this case swap was split, 2 each 1GB  partitions on USB flash plus 1 GB on the 
>>> microSD card. Previous attempts using 3 each 1 GB partitions on USB flash have been
>>> repeatedly successful, while a single attempt using 3 each 1GB partitions on microSD
>>> failed.
>> 
>> The more swap partitions (or space?) not on a /dev/mmcsd0s* the less of the
>> activity that /dev/mmcsd0 handles and likely the more time it tends to have
>> between explicit operations to do internal housekeeping before the next
>> explicit operation.
>> 
>> So the better approximation to not using /dev/mmcsd0 at all might not be
>> all that much of a surprise at having less of a problem on the device
>> (or a problem less often).
>> 
> In earlier tests (same card type) putting _all_ swap on microSD (along with /tmp) 
> avoided OOMA kills. Dependable mischief seems to come when swap is on both microSD 
> and USB.

As I do not know what all is going on for Out Of Memory kills, I was only
directly talking of the likes of ms/w and/or ms/r time figures being large
sometimes. OOMA may have other issues contributing to some of its activity
for all I know.

Is a difference, for example, 2 GB total swap vs. 3 GB total swap and the
message about the swap configuration? The memory tracking swap can fragment
and be a source of not being able to use swap as I understand. (I've quoted
the man page material in the past in these exchanges.) Large swap spaces
can hit such fragmentation issues sooner. (So there is a tradeoff in
having more swap than needed as the amount-more increases.)

Trev has reported: "I have run the -j4 buildworld now 11 times using a single
2G partition". I wonder what would happen for him if he used two 1 GB
partitions or three about 2/3 GB byte partitions. The contrasting case of a
working environment changed to have multiple partitions (with the same total)
could indicate something about if multiple partitions are sufficient to lead
to problems (ms/w, ms/r, OOM killing, or some combination).

> It's understood that USB and Ethernet share I/O hardware, but I thought microSD 
> was at least somewhat independent. Is this wrong?
> 

This I do not know.


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)