Ceph Storage question

[ilia987]

I am looking for a storage solution for scratchpad (need a fast storage, shared over all the cluster,can be unreliable. the data is used only data computational tasks) and after the computation the data is deleted

currently i have the following setup:

• proxmox cluster consist of 10 nodes
• ceph storage ssd based (over 3 nodes, will be scaled to 5 node to increase volume and redundancy )
• freenass storage (will be scrapped in the near future) with existing scratchpad storage (pcie-nvme, with no redundancy),
• in the near future 5 more nodes with 60HDD will be added , to create another ceph pool for low IOPS depend and high capacity data)

what would be the best approach to make a fast scratch pad, a new ceph pull with replication 1 with 3-5 osd (pcie cards)?
or to get a dedicated server and make a raid for them ?

 

[Alwin]

what would be the best approach to make a fast scratch pad, a new ceph pull with replication 1 with 3-5 osd (pcie cards)?
or to get a dedicated server and make a raid for them ?

Both have their dis-/advantages. NVMe will need a lot of interrupt handing. Ceph distributes that, but all nodes need enough resources to handle. While you can scale the dedicated hardware of one node on the beginning.

 

[ilia987]

what is going to happen for the pool (replication 1) when one of the osd dies?
the data is not important (it just scratch\tmp ) , but it is required that the pool will up and running for future tasks

 

[Alwin]

Nothing happens with the pool. The data of the dead OSD will be recovered on another OSD. Hence, 3x replica.

 

[ilia987]

Nothing happens with the pool. The data of the dead OSD will be recovered on another OSD. Hence, 3x replica.

my question was on replica 1 . (not on replica 3)
i am looking for a solution for fast scalable storage where no redundancy is needed (after the task is done all the data inside no longer needed)