I replaced the SSD for ZFS log by "Samsung SM883 240GB 2.5in Data Center SATA 6Gbps"
Seq. Read 540 MB/s Seq. Write 480 MB/s
Ran. Read 97K IOPS Ran. Write 22K IOPS
Not really a big improvement...
Uploaded 110 chunks in 5 seconds.
Time per request: 46514 microseconds...
indeed, the performance of the Crucial is not very good in writing:
Sustained Sequential Write up to (128k transfer) 95MB/s
Sustained Sequential Read up to (128k transfer) 500MB/s
I read that we recommend a small capacity SSD drive, and I took the 1st small size drive that I found...
I will reconfigure my disks in Raid10 and do some tests.
For the CPU .. are you sure AES is not supported? Intel says this:
Intel® AES New Instructions Yes
For the SSD log, I did not see any improvement with it, and when I monitor with iostat, I do not see any activity.
I just set up a 1st PBS and I have very poor performance:
Uploaded 99 chunks in 5 seconds.
Time per request: 51253 microseconds.
TLS speed: 81.83 MB/s
SHA256 speed: 188.74 MB/s
Compression speed: 317.88 MB/s
Decompress speed: 582.48 MB/s
AES256/GCM speed: 112.44 MB/s
Ok I think I figured out how to do it.
I deleted the "storage0" datastore which pointed to the ZFS path "/mnt/datastore/storage0"
and I created a datastore "pve1" to /mnt/datastore/storage0/pve1 and datastore "pve2" /mnt/datastore/storage0/pve2
is that the right way to go?
Hope this question has not been addressed yet, I checked before but found nothing.
I have 2 PVE and a 1 PBS with 1 datastore.
I connected 2 PVEs to the PBS.
The 1st PVE is mine, the 2nd is not mine.
PVE 1 sees the backups of 2.
PVE 2 sees the backups of 1.
Is it possible to isolate...
I do not understand because the quorum detected my node "down" rather quickly.
But the HA is really slow to launch the VM/CT on another node. I searched everywhere on the side of HA, and I do not see any modifiable value.
I just realized a test lab with 3x Proxmox 5.4 with Ceph and HA.
I configured watchdog on the IPMI on 10 seconds.
In spite of that, if I disconnect a node, the other 2 take almost 3 minutes to see it and restart the VM.
Is it possible to optimize this time to less than 1 minute ?