Hello :
I recently encountered a problem with random read iops when performing performance tests.
The scene I built is: pve5.4 + ceph 12.2.12
3 hosts, each host 1ssd + 3hdd (ssd is used for db and wal, hdd is osd)
KVM's disk will enable cache = write back mode, and select VirtIO SCSI single
Zh
Delay in testing random reads, iodeph = 1, numjobs = 1, the average read iops is 200+
As shown below:
Delay in testing random reads, iodeph = 2, numjobs = 1, average read iops is 500+
As shown below:
Delay in testing random reads, iodeph = 1, numjobs = 2, the average read iops is 500+
As shown below:
After testing, it is found that single io and single-threaded iops are fixed at a certain value. Is this normal? If it is not normal, how to adjust it to improve it? For example, adjusting the kvm configuration, the operating system or ceph itself?
thank a lot!
I recently encountered a problem with random read iops when performing performance tests.
The scene I built is: pve5.4 + ceph 12.2.12
3 hosts, each host 1ssd + 3hdd (ssd is used for db and wal, hdd is osd)
KVM's disk will enable cache = write back mode, and select VirtIO SCSI single
Zh
Delay in testing random reads, iodeph = 1, numjobs = 1, the average read iops is 200+
As shown below:
Delay in testing random reads, iodeph = 2, numjobs = 1, average read iops is 500+
As shown below:
Delay in testing random reads, iodeph = 1, numjobs = 2, the average read iops is 500+
As shown below:
After testing, it is found that single io and single-threaded iops are fixed at a certain value. Is this normal? If it is not normal, how to adjust it to improve it? For example, adjusting the kvm configuration, the operating system or ceph itself?
thank a lot!