ericaltendorf / plotman

Chia plotting manager
Apache License 2.0
909 stars 280 forks source link

Lock tmp and dst together #616

Open syadnom opened 3 years ago

syadnom commented 3 years ago

I would like an option to lock the tmp and dst together. Chia will do a quick mv saving 108GB of write time. When paired with the archiver, that likely means a faster plot+archive that just the plot. 108GB of writing faster.

Also, for those of us using a number of disks, that keeps that write off the SATA bus which can help across the board.

I'm using this on 'manual' plots and it shaves at least 20-30 minutes off overall times for me.

syadnom commented 3 years ago

semi circular reference just for fun https://github.com/ericaltendorf/plotman/issues/578

ywarnier commented 3 years ago

I have the same request but translated into a different problem. I plot on 3 disks, which are also the dst disks (just separate directories), but when plotman finds them, it tries to copy the final file to a plot folder on another disk, which is already used to plot another plot.

Example from my plotman.yaml

tmp:
  - /mnt/chia1/temp
  - /mnt/chia2/temp
  - /mnt/chia3/temp
dst:
  - /mnt/chia1/plots
  - /mnt/chia2/plots
  - /mnt/chia3/plots

When I run plotman, it does something like this:

  #    plot id    k                   tmp                    dst    wall   phase    tmp     pid   stat      mem   user    sys     io                         
  0   49b2b49d   32        mnt/chia3/temp        mnt/chia1/plots    4:49     2:1   188G   27953    DSK     1.5G   4:45   0:17   0:21                         
  1   2aeeb333   32        mnt/chia2/temp        mnt/chia3/plots   12:13     2:5   233G    9117    RUN     1.5G   5:39   0:19   0:50                         
  2   ce348d38   32        mnt/chia1/temp        mnt/chia2/plots   12:46     4:0   109G    2982    DSK   260.7M   8:14   0:33   2:13 

So in the last case for example, it takes 12h46 instead of a manual 9h plot just because it's fighting against the compress process still going on in /mnt/chia2/temp ...

578 seems to be offering a solution for that, indeed. Thanks for the semi-circular reference :-)