Optimal mount options for Proxmox?

[Netizen]

Are those below the optimal mount options for Proxmox?
I see posts that claim ext3 is better,faster and some other posts contain options such as rw and realtime.
What are the recommended ones, from Proxmox staff?

---
cat /etc/fstab

proc /proc proc defaults 0 0
none /dev/pts devpts gid=5,mode=620 0 0
/dev/md/0 /boot ext3 defaults 0 0
# /dev/md/1 belongs to LVM volume group 'vg0'
# /dev/md/2 belongs to LVM volume group 'vg1'
/dev/vg0/swap swap swap defaults 0 0
/dev/vg0/root / ext4 defaults 0 0
/dev/vg0/backup /backup ext4 defaults 0 0
/dev/vg0/data /vz ext4 defaults 0 0
/dev/vg1/backup2 /opt ext4 defaults 0 0

---

Thank you

 

[tom]

a default 3.0 iso install shows the following.

cat /etc/fstab

# <file system> <mount point> <type> <options> <dump> <pass>
/dev/pve/root / ext3 errors=remount-ro 0 1
/dev/pve/data /var/lib/vz ext3 defaults 0 1
UUID=f1f8d7b5-1af2-47ce-a07e-a4a3327848a3 /boot ext3 defaults 0 1
/dev/pve/swap none swap sw 0 0
proc /proc proc defaults 0 0

 

[Netizen]

And I assume here that those should also be kept for the life of the server?

 

[tom]

I do not understand your question.

These are the default settings. If you change it, you should have a reason and deeper understanding for the changes. If you do have any real reason for other settings, use the defaults.

 

[Netizen]

No problems, its the language barrier. We basically mean the same thing.

Now, as I have not tested ext3 to compare it with that my ext4, do you think I should really re-install the server (already with a few VMs) or the ext4 is as good and it's not worth the trouble?

pveperf:
CPU BOGOMIPS: 52798.64
REGEX/SECOND: 1427022
HD SIZE: 19.84 GB (/dev/mapper/vg0-root)
BUFFERED READS: 135.89 MB/sec
AVERAGE SEEK TIME: 6.88 ms
FSYNCS/SECOND: 28.38
DNS EXT: 52.02 ms
DNS INT: 63.20 ms (mydomain.com)
--

2x1.5 TB SATA drives, on SOFTWARE Raid 1

 

[tom]

your fsync/second are too slow. this value should be at least 1000, otherwise you will see very bad performance.

btw, softraid is not recommended and not supported, see http://pve.proxmox.com/wiki/Software_RAID

 

[Netizen]

I see what you mean. But what it causing that? Is it JUST the fact that I am using SoftRaid1?
The server is this one>> http://www.hetzner.de/en/hosting/produkte_rootserver/ex6s

Is SoftRaid-1 enough to cause such slowness?

 

[tom]

I see what you mean. But what it causing that? Is it JUST the fact that I am using SoftRaid1?
The server is this one>> http://www.hetzner.de/en/hosting/produkte_rootserver/ex6s

Is SoftRaid-1 enough to cause such slowness?

together with a missing cache, yes.

If you want a trouble free system with great performance, use a recommended hardware:
http://www.proxmox.com/proxmox-ve/requirements

 

[mir]

your fsync/second are too slow. this value should be at least 1000, otherwise you will see very bad performance.

btw, softraid is not recommended and not supported, see http://pve.proxmox.com/wiki/Software_RAID

1000 is only a valid number using ext3. Below is the same test run on a server using ext4 but using a SSD disk.
CPU BOGOMIPS: 32547.06
REGEX/SECOND: 940527
HD SIZE: 9.18 GB (/dev/disk/by-uuid/f66cdc9c-9cf8-4c31-802d-21b9ffcf0493)
BUFFERED READS: 173.14 MB/sec
AVERAGE SEEK TIME: 0.27 ms
FSYNCS/SECOND: 273.76
DNS EXT: 57.29 ms
DNS INT: 0.72 ms (datanom.net)

My conclusion is that the pveperf test is not reliable.

 

[tom]

...
My conclusion is that the pveperf test is not reliable.

pveperf runs some simple tests (in a reliable way). why do you think your result is wrong?

but it is not a tool to simulate several IO loads on your storage stack. but discussing benchmarks tools is not the topic of this thread.

 

[mir]

It was your remark that FSYNCS/SECOND should be at least 1000 which was the cause of my post. Since pveperf does not give consistent readings on the same hardware if all other is equal except file system. I have run the test on identical hardware using different filesystems showing enormous different results in FSYNCS/SECOND while any other benchmark used did not show the same difference in result.

This is from a different node using ext3 but the disk is a SATA2 average platter disk:
CPU BOGOMIPS: 23999.08
REGEX/SECOND: 1273432
HD SIZE: 3.66 GB (/dev/mapper/pve-root)
BUFFERED READS: 75.02 MB/sec
AVERAGE SEEK TIME: 10.03 ms
FSYNCS/SECOND: 446.19
DNS EXT: 53.26 ms
DNS INT: 1.27 ms (datanom.net)

compare
FSYNCS/SECOND: 446.19
with a top SATA3 SSD using ext4:
FSYNCS/SECOND: 273.76

Do you honestly believe those readings are correct? (The same SSD disk scores over 1300 FSYNCS/SECOND using ext3 and the only tools given those numbers are pveperf. Any other benchmark tool gives better scores on this SSD disk using ext4)

 

[tom]

again, out of topic. please do not highjack threads and start discussing something different. pls open a new thread if you think there is an issue with pveperf.

 

[Netizen]

Do you honestly believe those readings are correct? (The same SSD disk scores over 1300 FSYNCS/SECOND using ext3 and the only tools given those numbers are pveperf. Any other benchmark tool gives better scores on this SSD disk using ext4)

Ok, a bit off topic but have you got an alternative method of benchmarking that we can use and is proven to be consistent?

 

[mir]

I'll start a new thread.

 

[dietmar]

Do you honestly believe those readings are correct? (The same SSD disk scores over 1300 FSYNCS/SECOND using ext3 and the only tools given those numbers are pveperf. Any other benchmark tool gives better scores on this SSD disk using ext4)

The pveperf bench is quite simple, so I believe those values unless you find a bug in the code. We use it about 5 year now an it turns out to be a fery good indicator for performance problems (which is the purpose of that tool).