Hi- I managed to improve my throughput somewhat by recreating the RBD Image with a larger object size (I chose 16 Mb, not from any science but on a gut feel ) and then change the stripe to 4 This seemed to roughly double the performance from an average of 8-12Mb/s write to 16-24 Mb/s. I also changed the pool to PG Autoscale, and it has spent the past day or so reducing the number of PGs to what appears to be a target of 64. I am now seeing higher write speeds of 40Mb/s to as high as 100Mb/s. The average response time shown in windows task manager appears to be 'off' - I have seen it jump around from 500 ms to over 300 seconds back to 3 seconds in a matter of a few refreshes. Windows resource monitor is showing a more consistent response time on multiple parallel writes of about 1.5-3 seconds per write. -----Original Message----- From: Robert W. Eckert <rob@rob.eckert.name> Sent: Tuesday, May 7, 2024 8:36 AM To: ceph-users@ceph.io Subject: [ceph-users] Guidance on using large RBD volumes - NTFS Hi - in my home , I have been running cephfs for a few years, and have reasonably good performance, however since exposing cephfs via SMB has been hit and miss. So I thought I could carve out space for a RBD device to share from a windows machine My set up: CEPH 18.2.2 deployed using ceph adm 4 servers running RHEL9 on AMD 5600g CPUs 64 Gb Each 10Gbe NICs 4x4Tb Hdd 1 2Tb NVME for DB/WAL rbd pool is set to auto PG - its currently at 256 I have tested the NIC connection between the servers and my PC, an each point to point works well at the 10Gbe speeds Now the problem I created a 8Tb RBD using rbd create winshare -size 8T -pool rbd rbd map winshare I prepped the drive, formatted it, and the drive appears cleanly as an 8 Tb drive. When I used fio on the drive/volume, speeds were good around 150-200 Mb/s. Then I started trying to populate the drive from a few different sources, and performance took a nose dive. - Write speeds are about 6-10 Mb/s, and windows task manager shows average response time anywhere from 500ms to 30 seconds. - mainly around 4 seconds. I don't see any obvious bottlenecks - cpu on the servers are about 5-10%, memory is good, Network is showing under 1 Gb/s on all servers. I am wondering if I needed to use different parameters for creating the volume? Or is there a practical limit to the volume size I exceeded? Thanks, Rob _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io