[PVE-User] LVM over ISCSI Bad performance

Sebastian Gruner s.gruner at procitec.de
Wed Jun 22 08:44:57 CEST 2016


Hi, you have to create a physical volume for each LUN

Pvcreate /dev/dm-0
Pvcreate /dev/dm-1

Then you can create a volumegroup

Vgcreate groupname /dev/dm-0

And extend the group with the second volume

Vgextend groupname /dev/dm-1

After that you can mount the lvm group aus storage in webinterface




###############################################################
Hi again^^,

I miss interpreted the benchmark result, with writeback cache is 5500 MB/s so thats ok... with direct IO it gets about 100 MB/s

But I have still no clue how I could use multipath with proxmox web interface.

regards,

Harald

Am 21.06.2016 um 19:15 schrieb Harald Leithner:
> Hi,
>
> I have a MSA2312 connected with 1gbit to a 3 node PVE 4.2 Cluster.
> I managed to get ISCSI and LVM working and attached it to a VM.
>
> Atm I didn't manage to get multipath working with PVE because the 
> Wikipage tells me how to create multipath (which works) but not how 
> PVE uses it for LVM VGs... maybe someone give me a hint.
>
> Anyway the main problem is that fio only gives me a read performance 
> of
> 5 mb/s if I attach a disk as virtio or scsi device with cache "Writeback".
>
> If I attach the same SAN Drive directly in the VM with iSCSI (with and 
> without multipath) I get about 112 MB/s.
>
> I also get this speed on the pve host.
>
> I use:
>
> fio --filename=/dev/sdb --direct=1 --rw=read --bs=1m --size=20G
> --numjobs=200 --runtime=60 --group_reporting --name=file1
>
> for benchmarking.
>
> While benchmarking the VM I see 1300% cpu usage on the host (mainly 
> sys
> time)
>
> Maybe someone has a idea?
>
> regards
>

--
Harald Leithner

ITronic
Wiedner Hauptstra├če 120/5.1, 1050 Wien, Austria
Tel: +43-1-545 0 604
Mobil: +43-699-123 78 4 78
Mail: leithner at itronic.at | itronic.at
_______________________________________________
pve-user mailing list
pve-user at pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user





More information about the pve-user mailing list