Memory limit of 3.5 GiB on 965GiB RAM device

[chourroutm]

Hi,
Following on seung-lab/cloud-volume#635, I've now switched to a Linux workstation with 1TiB RAM but I can't compute downscaled versions of a dataset with this > 3.5GiB error:

from taskqueue import LocalTaskQueue
import igneous.task_creation as tc
from pathlib import Path

output_dir = "output/mask.precomputed/"
output_dir = Path(output_dir)
mkdir(output_dir)

output_dir = output_dir.absolute().as_uri() + "/"

print(output_dir)

tq = LocalTaskQueue(parallel=True)
tasks = tc.create_downsampling_tasks(output_dir, mip=0, num_mips = 3, factor= (2,2,2), fill_missing=True, delete_black_uploads=True)
tq.insert(tasks)
tq.execute()
print("Done!")

file:///home/me/convert_with_cloudvolume/output/mask.precomputed/
WARNING: Memory limit (3500000000 bytes) too low to compute 3 mips at a time. 2 mips possible.
Volume Bounds:  Bbox([0, 0, 0],[1169, 1169, 1345], dtype=np.int32, unit='vx')
Selected ROI:   Bbox([0, 0, 0],[1169, 1169, 1345], dtype=np.int32, unit='vx')

This is the output of free -h:

               total        used        free      shared  buff/cache   available
Mem:           1.0Ti        26Gi       965Gi       1.0Gi        15Gi       974Gi
Swap:          476Gi          0B       476Gi

I've probably made a mistake on the line tq = LocalTaskQueue(parallel=True)?

[william-silversmith]

Hi! This will depend on your chunk size and data type. I suspect something is set a bit oddly, as usually 3.5 GB is enough for even very large data sets.

However, you can make individual tasks larger by setting memory_target=int(300e9) (300 GB for example)

[chourroutm]

Thanks, it did solve the issue!