On 26/11/2024 15:09, Peter Grandi wrote:
Regardless of the specifics: 4KiB write IOPS is definitely not what Ceph was designed for. Yet so many people know better and use Ceph for VM disk images, even with logs and databases on them.
It depends...for a single thread/queue depth Ceph will give roughly 3-5k iops read, 1k iops write. Traditional SANs can give 2-5x this iops, specially for writes, so they are much better if you have a low concurrency workload...but if you have a workload that requires hundreds or thousands of concurrent threads: like many vms, each with many applications, with many databases each doing many concurrent transactions, then Ceph makes sense as it will handle this concurrency, whereas your traditional SAN will saturate quickly and will not be able to give the nice performance numbers it had with qd=1.