Researcher is complaining about slow transfer rates from a mounted NFS share to the SSD path that is on a local NVMe disk on the GPU compute node. We tested transfer rates on the system using OS utilities and we get > 215MB/s. When the transfer is done by the CryoSparc Worker the researcher is seeing ~21 MB/s. Anything we can look at in CryoSparc from the sysadmin side to help out?
@jmnewton cryoSPARC uses shutil.copyfile() for transfers to cache. We do not expect this to cause a reduction in throughput as dramatic as you reported, but you may want to check performance using that python method, the relevant storage volumes and suitable test data outside cryoSPARC.