RPI3 swap experiments

[bob prohaska]

On Wed, Jul 04, 2018 at 07:09:49AM +1000, Trev wrote:
> Ok, the final, final swap experiment.
> 
> I created a 2G swap partition on the SDCard:
> 
> gpart show
> =>      63  31116225  mmcsd0  MBR  (15G)
>          63      2016          - free -  (1.0M)
>        2079    102400       1  fat32lba  [active]  (50M)
>      104479  31011809       2  freebsd  (15G)
> 
> =>       0  31011809  mmcsd0s2  BSD  (15G)
>           0  25165824         1  freebsd-ufs  (12G)
>    25165824   4194304         2  freebsd-swap  (2.0G)
>    29360128   1651681            - free -  (806M)
> 
> and, yes, it still died during make -j4 buildworld:
> 
> Jul  3 22:15:24 rpi3 kernel: pid 57011 (c++), uid 0, was killed: out of 
> swap space
> 
> (I forgot to start the logging script, so no stats).
> 
> Conclusion: The only way to buildworld on the RPi3B+ is to use an 
> external USB mechanical hard disk (my hybrid Seagate 500G 2.5" drive 
> required a powered USB hub). No issues on a RPi2B v1.1 with swap and 
> /usr on a USB memory key.

This is getting curiouser and curioser.

I just set up my Pi3 with new storage media using the June 28 snapshot.
Swap, /var, /tmp and /usr are on a new (different model number but same specs)
Sandisk Extreme Go 64 GB USB flash drive. By all history, buildworld should have
failed, but didn't. The system updated itself to r335835 without a hitch.

However, running Peter Holms' stress2 suite using the all.sh script did, after
fifteen hours on r335835, produce a crash that superficially looks similar to 
those seen earlier (g_vfs_done(), (da0:umass-sim0:0:0:0):, vm_fault:) ending 
with
Stopped at      cluster_write+0x228:    ldr     x9, [x8, #80]!

It's tempting to attribute the difference in behavior  to the different USB
flash drives.

The one that gave the most trouble is labeled
Sandisk Extreme Go USB 3.1 model # SDCZ800-064G
The one in use now, which seems better-behaved, is labeled
Sandisk Extreme USB 3.0 model SDCZ80-064G 
and is some months older.

The practical difference is that the Extreme Go  model has replaced 
the ordinary Extreme, which is now scarce.

It looks as if FreeBSD distinguishes them by name but not quirk:

The troublemaker (Extreme G USB 3.1) is recognized as
da0 at umass-sim0 bus 0 scbus0 target 0 lun 0
da0: <SanDisk Extreme 1.00> Removable Direct Access SPC-4 SCSI device
da0: Serial Number 4C530001211014123270
da0: 40.000MB/s transfers
da0: 59840MB (122552320 512 byte sectors)
da0: quirks=0x2<NO_6_BYTE>

while the one that works better (Extreme USB 3.0) reports
da1 at umass-sim1 bus 1 scbus1 target 0 lun 0
da1: <SanDisk Extreme 0001> Removable Direct Access SPC-4 SCSI device
da1: Serial Number AA010428162242131598
da1: 40.000MB/s transfers
da1: 59836MB (122544516 512 byte sectors)
da1: quirks=0x2<NO_6_BYTE>


Details are at
http://www.zefox.net/~fbsd/rpi3/swaptests/r335835/

Thanks for reading!

bob prohaska