Proxmox really slow

[np86]

I decided to try out Proxmox (again). Since I am tired of vcenter`s high demand for memory

But my vms seems pretty slow, and unresponsive at times, the same goes for the webui.

My windows vm even stopped responding while i was converting a image to qcow.

I am also getting this message when i am shutting down or rebooting my container:

kernel:[10730.464939] unregister_netdevice: waiting for lo to become free. Usage count = 1


Anything special I need to do?

I am on proxmox 5 beta 2 at the moment, but same scenario at 4.4

Storage:
2x 250GB Samsung Evo 850 in raid1 on a 9211-8i with EXT4.

Server:
E5-2620 V4, 32GB ECC, Supermicro X10SRA-F

#pveperf
CPU BOGOMIPS: 67039.04
REGEX/SECOND: 2725977
HD SIZE: 219.88 GB (/dev/mapper/pve-root)
BUFFERED READS: 196.10 MB/sec
AVERAGE SEEK TIME: 0.25 ms
FSYNCS/SECOND: 279.51
DNS EXT: 47.66 ms
DNS INT: 18.74 ms

EDIT: I also tried using my onboard controller with ZFS raid1. (Version 4.4)
Seems to be the same issue, under qemu-convert.

 

[talos]

You have really low fsyncs. Windows use sync writes all over the place and get a big hit if it has to wait for data to be written. Samsung Consumer SSDs are a really bad choice for Virtualization and Servers in general.

Here is one of my private (old) Server. ZFS Raid10 4x8TB Seagate Enterprise NAS HDD with an Intel DC3700S 100GB as Log (for caching Sync writes).

# pveperf /datapool1/media
CPU BOGOMIPS: 54399.76
REGEX/SECOND: 2226463
HD SIZE: 13133.76 GB (datapool1/templates)
FSYNCS/SECOND: 3119.04
DNS EXT: 17.37 ms
DNS INT: 9.21 ms

The log SSD bumped my FSYNC from around 600 to almost 3200. I see a lot of action on the SSD if using Windows (general Usage) or Databases under Linux.

I recommend use Intel Datacenter SSDs. This SSDs have a capacitor build it to protect the SSD Memory Cache from a power outtage. This SSD always reports data written to the OS, even if it is still in the memory cache of the SSD. That makes sync writes really fast.

 

[hieve]

I've been using software raid on my productive system and It's really really bad with perfomance and I/O Delay.
I tried using owncloud as VM and one sync on the owncloud vm destroyed every others VM perfomance.
Now without Software Raid on my test system the perfomance is much better and there are much more capabilities for more VMs without huge impacts.

I would recommended trying Hardware Raid or first switching the drives for others

 

[fortechitsolutions]

Just to chime in to agree, this seems a 'storage performance' sort of issue as the fsyncs/s stat suggests from your CLI.

things that would help
- I like the suggestion of the intel SSD, probably is simplest
- more 'splindles' will increase your performance here, ie, more parallel channels for pushing IOPS. in your case more SSDs
- invest in a modest price HW Raid card, is probably simplest fix? (ie, a ~$200ish LSI-raid will let you do raid1/10 and get better overall performance I believe; or there are other vendors well supported; possibly others will chime in as to their favoured part(s)
- or, just don't worry, since you know the cause of the 'problem'. If you don't mind the IOPS performance restriction, I am sure otherwise it works pretty smoothly. (?)

I would not expect ZFS to magically fix things with same hardware config. In my experience (albeit limited) with ZFS, you will need to have 'plenty of spindles' (parallel channels for IOPS data to be written/read in parallel) and 'cache volumes' on SSD ideally (ZIL etc) to help.

Tim

 

[np86]

I tried my SSD`s in a HP server with a P440 controller. Performance was a bit better, but still high I/O.

What Enterprise SSDs are recommended? I have been on the lookout for HGST and Intel S3700 so far. But I am not sure if they would be the best fit for the price