Performance problem on one vmhost proxmox 3.1

[ewuewu]

Hello Forum,

we have a strange performance problem running a VM on our 3 node cluster.
The vm is a a windows 2003 Standard edition with no virtio driver.
Once a day there is a high perfomce database job running on this vm.
If we run this vm on our different proxmox hosts we discovered that the newest host delivers the worst performance.

We achieve the following job runtimes on our different proxmox hosts:
Host A: (6 years old) 44 minutes
Host B (3 years old) 38 minutes
Host C (brandnew) 68 minutes

All hosts are running with an actual proxmox 3.1. They are all connected via dedicated bonded nic to a nfs storage.

The hosts are consisting of the following hardware:

Host A:

Intel(R) Xeon(R) CPU E5420 @ 2.50GHz
32 GB RAM
Supermicro X7DB8 (Bios Phoenix Vers. 2.1c)
3ware 9650SE SATA-II RAID
One Gigabit ET Dual Port Server Adapter 82576 Gigabit Network Connection (igb / bonded)
Two Onboard Intel 80003ES2LAN Gigabit Ethernet Controller (Copper) (igb / bonded)
Proxmox is installed on hardware raid1 with two SATA Disks

~# pveperf
CPU BOGOMIPS: 39996.92
REGEX/SECOND: 988985
HD SIZE: 73.33 GB (/dev/mapper/pve-root)
BUFFERED READS: 112.24 MB/sec
AVERAGE SEEK TIME: 9.84 ms
FSYNCS/SECOND: 1723.32
DNS EXT: 103.80 ms
DNS INT: 2949.81 ms

Host B:

Intel "Intel(R) Xeon(R) CPU X5650 @ 2.67GHz
64 GB RAM
Adaptec ASR5805 (aacraid driver)
82576 Gigabit ET Dual Port Server Adapter / 82576 Gigabit Network Connection (igb driver / bonded)
Onboard: Two Intel 82576 Gigabit Network Connection (igb driver / bonded)
Proxmox is installed on hardware raid1 with two SAS disks

pveperf
CPU BOGOMIPS: 127995.72
REGEX/SECOND: 1093625
HD SIZE: 33.47 GB (/dev/mapper/pve-root)
BUFFERED READS: 184.50 MB/sec
AVERAGE SEEK TIME: 4.19 ms
FSYNCS/SECOND: 3302.67
DNS EXT: 57.02 ms
DNS INT: 2858.06 ms

Host C:

Intel(R) Xeon(R) CPU E5-2640 0 @ 2.50GHz
64 GB RAM
Intel S2600CP (Bios: Intel SE5C600.86B.01.08.0003.022620131521)
Adaptec 7805Q MaxC 8Port
2 x Intel Dualport Server Adapter (bonded)
2 x Intel Onboard NIC (bonded)
Proxmox is installed on hardware raid1 with two SSD disks

# pveperf
CPU BOGOMIPS: 119705.52
REGEX/SECOND: 1176046
HD SIZE: 41.09 GB (/dev/mapper/pve-root)
BUFFERED READS: 729.30 MB/sec
AVERAGE SEEK TIME: 0.23 ms
FSYNCS/SECOND: 2738.21
DNS EXT: 51.01 ms
DNS INT: 3162.55 ms

What is the best method to find the obviously existing bottleneck on Host C
Any Help is appreciated

 

[udo]

Hello Forum,

we have a strange performance problem running a VM on our 3 node cluster.
The vm is a a windows 2003 Standard edition with no virtio driver.
Once a day there is a high perfomce database job running on this vm.
If we run this vm on our different proxmox hosts we discovered that the newest host delivers the worst performance.
...

What is the best method to find the obviously existing bottleneck on Host C
Any Help is appreciated

Hi,
I guess it's an IO-problem because of heavy IO during database-job?!

Look during runtime with top for wait (io) and cpu usage.
Try to find the bottleneck (network due nfs?!) with iperf.
Compare the VM on host-C with local storage.

Look with pveperf on your nfs-storage (like pveperf /mnt/nfs-share ).
Perhaps something wrong with bonding/MTU?

Udo