[ovirt-users] Ovirt & Ceph

[Alessandro De Salvo]

Hi Yaniv,

> Il giorno 18 dic 2016, alle ore 17:37, Yaniv Kaul <ykaul at redhat.com> ha scritto:
> 
> 
> 
>> On Sun, Dec 18, 2016 at 6:21 PM, Alessandro De Salvo <Alessandro.DeSalvo at roma1.infn.it> wrote:
>> Hi,
>> having a 3-node ceph cluster is the bare minimum you can have to make it working, unless you want to have just a replica-2 mode, which is not safe.
> 
> How well does it perform?

One if the ceph clusters we use had exactly this setup: 3 DELL R630 (ceph jewel), 6 1TB NL-SAS disks so 3 mons, 6 osds. We bound the cluster network to a dedicated interface, 1Gbps. I can say it works pretty well, the performance reaches up to 100MB/s per rbd device, which is the expected maximum for the network connection. Resiliency is also pretty good, we can loose 2 osds (I.e. a full machine) without impacting on the performance.

>  
>> It's not true that ceph is not easy to configure, you might use very easily ceph-deploy, have puppet configuring it or even run it in containers. Using docker is in fact the easiest solution, it really requires 10 minutes to make a cluster up. I've tried it both with jewel (official containers) and kraken (custom containers), and it works pretty well.
> 
> This could be a great blog post in ovirt.org site - care to write something describing the configuration and setup?

Oh sure, if it may be of general interest I'll be glad to. How can I do it? :-)
Cheers,

   Alessandro 

> Y.
>  
>> The real problem is not creating and configuring a ceph cluster, but using it from ovirt, as it requires cinder, i.e. a minimal setup of openstack. We have it and it's working pretty well, but it requires some work. For your reference we have cinder running on an ovirt VM using gluster.
>> Cheers,
>> 
>>    Alessandro 
>> 
>>> Il giorno 18 dic 2016, alle ore 17:07, Yaniv Kaul <ykaul at redhat.com> ha scritto:
>>> 
>>> 
>>> 
>>>> On Sun, Dec 18, 2016 at 3:29 PM, rajatjpatel <rajatjpatel at gmail.com> wrote:
>>>> ​Dear Team,
>>>> 
>>>> We are using Ovirt 4.0 for POC what we are doing I want to check with all Guru's Ovirt.
>>>> 
>>>> We have 2 hp proliant dl 380 with 500GB SAS & 1TB *4 SAS Disk and 500GB SSD.
>>>> 
>>>> Waht we are done we have install ovirt hyp on these h/w and we have physical server where we are running our manager for ovirt. For ovirt hyp we are using only one 500GB of one HDD rest we have kept for ceph, so we have 3 node as guest running on ovirt and for ceph. My question you all is what I am doing is right or wrong.
>>> 
>>> I think Ceph requires a lot more resources than above. It's also a bit more challenging to configure. I would highly recommend a 3-node cluster with Gluster.
>>> Y.
>>>  
>>>> 
>>>> Regards
>>>> Rajat​
>>>> 
>>>> 
>>>> _______________________________________________
>>>> Users mailing list
>>>> Users at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/users
>>>> 
>>> 
>>> _______________________________________________
>>> Users mailing list
>>> Users at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/users
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20161218/5c88bc4d/attachment.html>