Re: [Users] Virtual Disk creation - really slow (config issue?) e.g 20GB disk created in 23 mins - NFS server/Gigabit switch - can anyone suggest why speed is so slow ? (at this rate it would take 1hr 40 mins to create a 100GB disk...)
Thank you Gary. After re-mounting the NFS share with async it dramatically increased the speed ! I can now create a 10GB virtual disk in 2 mins (rather than 10 mins) Is using the async option more likely to cause data corruption though ? Many thanks Regards On 22 March 2012 21:54, Gary Scarborough <virtuallymad@gmail.com> wrote:
On Thu, Mar 22, 2012 at 12:59 PM, Morgan Cox <morgancoxuk@gmail.com>wrote:
Hi.
After my many issues with spice I have finally sorted them out.
Now I am having an issue with the speed of virtual disk creation - when I go to create a virtual disk it takes a long time (far to long to be usable)
My setup is as following
3 servers:-
1 Frontend (ovirt engine) - Fedora 16 - external + local IP (91.215.xxx.xx + 10.0.0.10) 1 Ovirt node (using the node .iso) - external + local IP (ovirtmgmt (eth0) - 91.215.xxx.xxx and LAN (eth1) - 10.0.0.101) 1 NFS storage - external + local IP - I attached the storage using the Local IP (10.0.0.190)
All servers are connected via a Gigabit switch (so speed should be better)
I can scp between servers far faster - I'm almost sure its due to the node ovirtmgmt address being an external (internet ip) and not LAN ip. - therefore the traffic is not just going over the gigabit switch but via the core router in the DC...
Can anyone suggest how I can 'debug' where the slowness is coming from
On my NFS server I have it setup like (in /etc/exports)
/storage1 *(rw,sync,no_subtree_check)
I had previous tried setting up the node on a LAN IP previously and it caused spice to not work - as the node had no internet access / DNS - I also tried making the frontend server a gateway (and using - iptables -t nat -A POSTROUTING -j MASQUERADE ) - this also cause spice to not work (as it always tried to connect to the 10.0.0.x address from my desktop - and fail..)
Any hints/help will be welcomed
Cheers
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
You could try async instead of sync. You could also try thin provisioning.
-- Gary Scarborough IST Lab Manager Rochester Institute of Technology Rochester NY
It can. I have seldom had need to worry about it. Most data centers have battery backups to keep power outages from crashing systems. I am also not sure if its any more or less of a risk than iscsi, except you may not be able to tell right away. async works by basically lying to the client about the read/write. Keep in mind that the read/write may be into a virtual machine in this case. Even if all packets are accounted for, not completing the full stream of writing could lead to a corrupt VM filesystem anyway. In my mind, its worth the risk for the speed. This is also why we do backups. On Fri, Mar 23, 2012 at 7:48 AM, Morgan Cox <morgancoxuk@gmail.com> wrote:
Thank you Gary.
After re-mounting the NFS share with async it dramatically increased the speed !
I can now create a 10GB virtual disk in 2 mins (rather than 10 mins)
Is using the async option more likely to cause data corruption though ?
Many thanks
Regards
On 22 March 2012 21:54, Gary Scarborough <virtuallymad@gmail.com> wrote:
On Thu, Mar 22, 2012 at 12:59 PM, Morgan Cox <morgancoxuk@gmail.com>wrote:
Hi.
After my many issues with spice I have finally sorted them out.
Now I am having an issue with the speed of virtual disk creation - when I go to create a virtual disk it takes a long time (far to long to be usable)
My setup is as following
3 servers:-
1 Frontend (ovirt engine) - Fedora 16 - external + local IP (91.215.xxx.xx + 10.0.0.10) 1 Ovirt node (using the node .iso) - external + local IP (ovirtmgmt (eth0) - 91.215.xxx.xxx and LAN (eth1) - 10.0.0.101) 1 NFS storage - external + local IP - I attached the storage using the Local IP (10.0.0.190)
All servers are connected via a Gigabit switch (so speed should be better)
I can scp between servers far faster - I'm almost sure its due to the node ovirtmgmt address being an external (internet ip) and not LAN ip. - therefore the traffic is not just going over the gigabit switch but via the core router in the DC...
Can anyone suggest how I can 'debug' where the slowness is coming from
On my NFS server I have it setup like (in /etc/exports)
/storage1 *(rw,sync,no_subtree_check)
I had previous tried setting up the node on a LAN IP previously and it caused spice to not work - as the node had no internet access / DNS - I also tried making the frontend server a gateway (and using - iptables -t nat -A POSTROUTING -j MASQUERADE ) - this also cause spice to not work (as it always tried to connect to the 10.0.0.x address from my desktop - and fail..)
Any hints/help will be welcomed
Cheers
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
You could try async instead of sync. You could also try thin provisioning.
-- Gary Scarborough IST Lab Manager Rochester Institute of Technology Rochester NY
-- Gary Scarborough IST Lab Manager Rochester Institute of Technology Rochester NY
Many thanks Gary On 25 March 2012 02:25, Gary Scarborough <virtuallymad@gmail.com> wrote:
It can. I have seldom had need to worry about it. Most data centers have battery backups to keep power outages from crashing systems. I am also not sure if its any more or less of a risk than iscsi, except you may not be able to tell right away. async works by basically lying to the client about the read/write. Keep in mind that the read/write may be into a virtual machine in this case. Even if all packets are accounted for, not completing the full stream of writing could lead to a corrupt VM filesystem anyway. In my mind, its worth the risk for the speed. This is also why we do backups.
On Fri, Mar 23, 2012 at 7:48 AM, Morgan Cox <morgancoxuk@gmail.com> wrote:
Thank you Gary.
After re-mounting the NFS share with async it dramatically increased the speed !
I can now create a 10GB virtual disk in 2 mins (rather than 10 mins)
Is using the async option more likely to cause data corruption though ?
Many thanks
Regards
On 22 March 2012 21:54, Gary Scarborough <virtuallymad@gmail.com> wrote:
On Thu, Mar 22, 2012 at 12:59 PM, Morgan Cox <morgancoxuk@gmail.com>wrote:
Hi.
After my many issues with spice I have finally sorted them out.
Now I am having an issue with the speed of virtual disk creation - when I go to create a virtual disk it takes a long time (far to long to be usable)
My setup is as following
3 servers:-
1 Frontend (ovirt engine) - Fedora 16 - external + local IP (91.215.xxx.xx + 10.0.0.10) 1 Ovirt node (using the node .iso) - external + local IP (ovirtmgmt (eth0) - 91.215.xxx.xxx and LAN (eth1) - 10.0.0.101) 1 NFS storage - external + local IP - I attached the storage using the Local IP (10.0.0.190)
All servers are connected via a Gigabit switch (so speed should be better)
I can scp between servers far faster - I'm almost sure its due to the node ovirtmgmt address being an external (internet ip) and not LAN ip. - therefore the traffic is not just going over the gigabit switch but via the core router in the DC...
Can anyone suggest how I can 'debug' where the slowness is coming from
On my NFS server I have it setup like (in /etc/exports)
/storage1 *(rw,sync,no_subtree_check)
I had previous tried setting up the node on a LAN IP previously and it caused spice to not work - as the node had no internet access / DNS - I also tried making the frontend server a gateway (and using - iptables -t nat -A POSTROUTING -j MASQUERADE ) - this also cause spice to not work (as it always tried to connect to the 10.0.0.x address from my desktop - and fail..)
Any hints/help will be welcomed
Cheers
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
You could try async instead of sync. You could also try thin provisioning.
-- Gary Scarborough IST Lab Manager Rochester Institute of Technology Rochester NY
-- Gary Scarborough IST Lab Manager Rochester Institute of Technology Rochester NY
participants (2)
-
Gary Scarborough -
Morgan Cox