RPI3 swap experiments

[bob prohaska]

On Wed, Jul 18, 2018 at 12:15:52PM +0100, tech-lists wrote:
> 
> I've not read the whole thread yet, but have been following parts of it 
> with interest. I'm just writing to say that I'm right now running [1] a 
> rpi3 (the earlier one, not the 3B+) on -current [2] and am having no 
> issues at all with swap.
> 
I don't mean to suggest that swap can't be made to work; only that the error
messages presented  seem inconsistent with what's actually going wrong.

[snip]
 
> I think what you're running into when encountering these errors is 
> firstly the disk erroring and then, because writes are too slow for the 
> usage, these errors amplifying. But that being a function of the 
> slowness of combined usb2 interface/mmc/flash. The flash erroring might 
> be in spec for flash in this context.
>
In the example at
http://www.zefox.net/~fbsd/rpi3/swaptests/r336356/2gbusbflash_1gbsdflash/
The greatest delay in write for swap is 368.0 ms/w, at timestamp Tue Jul 17 18:47:06 PDT 2018
The OOMA kill appears to happen much later, at Tue Jul 17 20:38:58 PDT 2018
when a read delay for microSD swap hits 263.6 ms/r

At best, the error message is misleading. Being out of swap and having to wait 
longer than usual to access it are very different problems. 
 
It seems to be an article of faith that flash is "too slow". I do agree that FreeBSD
deals more reliably with mechanical disks, but given BSD history that should come as 
no surprise 8-) Use of flash storage has grown, it is unlikely to stop. 

It would be very interesting if the OOMA kill trigger could be adjusted or turned off.

Thanks for reading,

bob prohaska