GPT vs MBR for swap devices

[bob prohaska]

On Fri, Jun 15, 2018 at 09:57:17AM -0600, Warner Losh wrote:
> 
> Right, this is why I was suggesting gstat. It's a direct measure of the
> read/write performance of the device with some latency numbers. It will
> give the kind of data I'm looking for. 

Here is a first cut at recording gstat data on r334939 during two -j4 buildworld 
sessions.  The interesting one uses 2 GB of SD flash swap on the boot device and 
1 GB of USB flash on the thumb drive hosting the outboard file systems. The readme
and log file are at
http://www.zefox.net/~fbsd/rpi3/swaptests/

The logging command was
gstat -aBd -I 10s > 2gbsdflash_1gbusbflash.log

It looks as if something strange happens when swap is placed on two separate
interfaces. In the mixed USB/SD case, it appears that the SD flash swap is ignored
and the USB flash swap is never very busy (eyeball scan only). 

The failed buildworld ended at the bootstrap tools stage, swap was on
da0b and mmcsd0s3d. 

It's already known that having all swap on SD flash in two partitions or 
all swap on USB mechanical disk in one partition works. There is also an 
RPI2 case with all swap on USB flash in one partition at
http://www.zefox.net/~fbsd/rpi2/swaptests/ in case it's of interest.

At this stage I don't know what to try next, suggestions are welcome.
It would be easy to sample at 1 second rather than ten. Cutting SD flash
swap down to 1 or 2 GB is also easy. Likewise, reducing to 1 GB USB
flash swap is easy. I have a powered hub so USB things can be added, 
but that makes the setup more complicated rather than less.



Thanks very much for reading,

bob prohaska