Recent content by michaeljk

Hi, vielleicht helfen dir die folgenden Infos weiter: https://www.hardwareluxx.de/community/threads/proxmox-stammtisch.1039627/post-28522188 https://github.com/docker/for-linux/issues/711

Durch die aktivierten Caches dürfte man aber einen guten Teil der Sicherheit verlieren, den ZFS eigentlich bieten soll. Sofern der Controller meldet, dass die Daten bereits weggeschrieben wurden, sich aber noch in irgendeinem Cache im System befinden, so wäre dies bei einem Absturz oder...

Ceph über WAN mit Gigabit-Anbindung? Halte ich für keine gute Idee, selbst im lokalen LAN sollte das Storage-Netzwerk schneller angebunden sein :) Was genau hast du denn vor, weil du die Nodes über mehrere Rechenzentren hinweg auftrennen möchtest? Ausfallsicherheit / Backup-Standort? Dann würde...

@thoand: Booten von einem ZFS-Mirror (RAID1) von NVMe ist ab Proxmox 6 kein Problem mehr, selbst getestet, funktioniert. In Verbindung mit ZFS solltest du allerdings noch auf folgende Dinge achten: - Keinen Hardware-RAID Controller benutzen - Je nach Anzahl / RAM-Bedarf der virtuellen Maschinen...

Kurze Info: Heute die Installation auf einem EX42-NVME mit der Proxmox 6 Beta getestet, dort funktioniert der Bootvorgang dann völlig problemlos. Die Performance ist allerdings niedriger als erwartet: pveperf /rpool/data CPU BOGOMIPS: 57600.00 REGEX/SECOND: 4497862 HD SIZE...

I just upgraded the Proxmox host from 5.2 to 5.3-5. Mostly standard configuration on a Dell T20 machine, ZFS with mirrored harddisks. Update worked fine, except that the only Windows 10 VM is now incredibly slow as soon as it gets to the login screen. There's no high CPU usage or I/O, it just...

We used some Gen6 machines on our old cluster some time ago, they had the some P420i controller - no problems, worked perfectly with SAS and SATA drives with Raid 10 configured.

Just a short reply for this issue: KSM_TUNED didn't fixed it, so I gave up on this and moved the virtual machines to a second host (HP Gen9 with hardware raid and LVM-thin installed). Now everything works as expected, I/O is almost always below 1%. I guess that there is some sort of hardware...

cat /etc/default/ksmtuned # Defaults for ksmtuned initscript # sourced by /etc/init.d/ksmtuned # installed at /etc/default/ksmtuned by the maintainer scripts # # This is a POSIX shell fragment # # start ksmtuned at boot [yes|no] START=yes ps aux |grep ksm root 133 0.0 0.0 0...

cat /sys/kernel/mm/ksm/pages_sharing 0 So it seems that KSM is not beeing used at the moment because there's enough free memory available?

top - 15:20:45 up 3 days, 1:27, 2 users, load average: 5.81, 5.35, 5.20 Tasks: 1866 total, 1 running, 1865 sleeping, 0 stopped, 0 zombie %Cpu(s): 3.3 us, 1.3 sy, 0.0 ni, 81.0 id, 14.4 wa, 0.0 hi, 0.0 si, 0.0 st KiB Mem: 13191573+total, 13136350+used, 552236 free, 50927776...

In addition, io-wait from today: 06:44:41 PM CPU %user %nice %system %iowait %steal %idle 07:44:43 PM all 5.61 0.00 1.97 0.44 0.00 91.98 08:44:43 PM all 5.22 0.00 1.99 0.48 0.00 92.31 09:44:43 PM all...

After 24 hours, the problem is here again. Reboot was on friday, 15:30 and all looked good: Fri Sep 16 15:43:09 CEST 2016 total used free shared buffers cached Mem: 131915736 77955020 53960716 54944 22313508 102228 -/+ buffers/cache...

No entries in dmesg and CPU usage looked good - the high load value results mainly because of the high I/O wait. So it seems that at some time there's no more free memory. Currently it looks like this: uptime 17:17:21 up 3:24, 4 users, load average: 1.72, 1.67, 1.69 arcstat.py time...

We watched the new system for some more days. After a fresh reboot, iostat shows between 3-10% usage for each disk, overall 0.5-0.7 wait in top. These values increase over time, today 6 out of 8 HDDs showed an usage of 100% (!), the other 2 had 30-40%, overall 15-30 wait: (Reboot on 13:50...