I’m going well scaling horizontally (more machines) but having trouble scaling vertically on my highest performance machine, which is an i9 9900kf with 64GB of RAM and 2x 2TB NVME SSD drives.
It seems to be taking doing about 1 plot per hour (massively parallelised) ie about 2.5 tb / day, but I think (from seeing other’s outputs?) that it should be able to do more.
My question is, am I pushing this machine too hard? Would I likely get better performance by reducing the number of plots in P1?
Many thanks in advance for any suggestions
Here is the output from plotman:
plot id k tmp dst wall phase tmp pid stat mem user sys io
61bcb49b 32 /plots_tmp/0 /plots/1 0:04 1:2 44G 160409 SLP 7.6G 0:06 48s 0s
bfd04b47 32 /plots_tmp/1 /plots/2 2:09 1:5 174G 155224 SLP 7.6G 3:29 0:09 2s
99980f85 32 /plots_tmp/0 /plots/0 2:35 1:6 179G 153857 SLP 7.6G 4:11 0:11 4s
bc581f0d 32 /plots_tmp/1 /plots/1 3:06 1:7 175G 152354 SLP 7.4G 5:04 0:13 8s
bbb2c878 32 /plots_tmp/0 /plots/2 3:33 2:1 169G 150991 RUN 1.6G 5:38 0:15 45s
d5d24ae0 32 /plots_tmp/1 /plots/0 5:38 3:1 240G 146111 RUN 8.0G 7:34 0:19 0:03
16daaaf9 32 /plots_tmp/0 /plots/1 6:02 3:2 218G 144892 RUN 7.4G 7:56 0:21 0:04
770469b3 32 /plots_tmp/1 /plots/2 6:35 3:3 204G 143300 RUN 7.4G 8:23 0:23 0:06
c58f9f23 32 /plots_tmp/0 /plots/0 6:59 3:3 186G 142055 RUN 7.4G 8:46 0:24 0:07
cafc7bd3 32 /plots_tmp/1 /plots/2 9:06 3:6 169G 137030 RUN 7.4G 10:35 0:31 0:14
8852f5c3 32 /plots_tmp/0 /plots/1 9:26 4:0 109G 135970 RUN 555.6M 10:51 0:33 0:15
My plotman settings are:
tmpdir_stagger_phase_major: 2
tmpdir_stagger_phase_minor: 1
tmpdir_stagger_phase_limit: 2
tmpdir_max_jobs: 12
global_max_jobs: 20
global_stagger_m: 10
plotting:
k: 32
e: False
n_threads: 6
n_buckets: 128
job_buffer: 6500