Hello,
We run a PVE cluster of 5 nodes with ceph on each node.
Each node has a number of OSDs, each backed by SSDs of various sizes.
A few months ago the OSDs / SSD drives per node were as follows:
The cluster has only one ceph pool configured with the default 3 replicas and 512PGs.
With the above disk configuration the total usable disk space as calculated by ceph was 23.8TiB at the time.
Since the pool's utilization was getting close to 70+% we decided to add more disks.
On all five nodes we added an extra 6.99TiB SSD per node which amounted to 34.95TiB extra raw space and a theoretical 11.65TiB of extra usable space.
So afterwards the disk configuration became as follows:
But after ceph did its rebalancing the total usable space increased only by 4.3TiB, ie 28.1TiB total usable space.
Since we are using various sizes of disks, I suspect that the usable space might not be as simple as "raw disks / 3 replicas = usable space", but after installing the 5 new disks of same size I hoped that we it would have given us an increase of ~11TiB.
Obviously there's something I am missing in the way the usable space is calculated.
Could someone more experienced with ceph, shed some light as to why we only have 28.1TiB usable space over 162.73TiB of raw space, which divided by 3 it should theoretically give us ~54.24TiB of usable space?
I get that there are some losses due to overheads, metadata, etc, but we get almost half the usable space!
Would re-arranging the existing SSDs per node yield higher usable space?
If so, could someone provide some insight as to how we should go about it?
Let me know if you need me to provide you with any command results from ceph or PVE, if it will help you understand the situation better.
Thanks!
We run a PVE cluster of 5 nodes with ceph on each node.
Each node has a number of OSDs, each backed by SSDs of various sizes.
A few months ago the OSDs / SSD drives per node were as follows:
Code:
PVE1
4x 3.49TiB (3.84TB)
5x 1.75TiB (1.92TB)
3x 745GiB (800GB)
PVE2
4x 3.49TiB (3.84TB)
5x 1.75TiB (1.92TB)
3x 745GiB (800GB)
PVE3
4x 3.49TiB (3.84TB)
4x 1.75TiB (1.92TB)
6x 745GiB (800GB)
PVE4
2x 3.49TiB (3.84TB)
11x 1.75TiB (1.92TB)
PVE5
2x 3.49TiB (3.84TB)
11x 1.75TiB (1.92TB)
The cluster has only one ceph pool configured with the default 3 replicas and 512PGs.
With the above disk configuration the total usable disk space as calculated by ceph was 23.8TiB at the time.
Since the pool's utilization was getting close to 70+% we decided to add more disks.
On all five nodes we added an extra 6.99TiB SSD per node which amounted to 34.95TiB extra raw space and a theoretical 11.65TiB of extra usable space.
So afterwards the disk configuration became as follows:
Code:
PVE1
4x 3.49TiB (3.84TB)
5x 1.75TiB (1.92TB)
3x 745GiB (800GB)
1x 6.99TiB (7.68TB)
PVE2
4x 3.49TiB (3.84TB)
5x 1.75TiB (1.92TB)
3x 745GiB (800GB)
1x 6.99TiB (7.68TB)
PVE3
4x 3.49TiB (3.84TB)
4x 1.75TiB (1.92TB)
6x 745GiB (800GB)
1x 6.99TiB (7.68TB)
PVE4
2x 3.49TiB (3.84TB)
11x 1.75TiB (1.92TB)
1x 6.99TiB (7.68TB)
PVE5
2x 3.49TiB (3.84TB)
11x 1.75TiB (1.92TB)
1x 6.99TiB (7.68TB)
But after ceph did its rebalancing the total usable space increased only by 4.3TiB, ie 28.1TiB total usable space.
Since we are using various sizes of disks, I suspect that the usable space might not be as simple as "raw disks / 3 replicas = usable space", but after installing the 5 new disks of same size I hoped that we it would have given us an increase of ~11TiB.
Obviously there's something I am missing in the way the usable space is calculated.
Could someone more experienced with ceph, shed some light as to why we only have 28.1TiB usable space over 162.73TiB of raw space, which divided by 3 it should theoretically give us ~54.24TiB of usable space?
I get that there are some losses due to overheads, metadata, etc, but we get almost half the usable space!
Would re-arranging the existing SSDs per node yield higher usable space?
If so, could someone provide some insight as to how we should go about it?
Let me know if you need me to provide you with any command results from ceph or PVE, if it will help you understand the situation better.
Thanks!
Last edited: