Expanding storage: understanding "cloud" storage

[amil]

Hello,

With my current 3x 120 GB SSD hard disk configuration per dedicated server (single servers) -raid 5-, the server will become full in hard disk space very fast, while server is with low load (on cpu/memory). So this is a heavy inconvenient, since yes: we have a nice performance but too low space to use as local storage for VMs.

On this configuration, we have attached a NFS storage attached via Proxmox, so this storage is used as ISO/template storage and for backups. This storage is the "default" storage included with dedicated server, and it works well for it.

So in this point i think a good solution will be attach any kind of "cloud" storage to create here the VM instead on local hard disks.

Going practise, and due this dedicated servers are on OVH i am reading about their "cloud" storage options: i think is not a bad offer, and it suppossed that will work better, since it can be on same datacenter, but i am new in terms of new "cloud" storage and which method is better for every scenario, i ´ll try to explain.

Option A)

Object Storage - Incomming traffic free - Outgoing 0,01 €/per GB/month
[https://www.ovh.co.uk/cloud/storage/]

I see they use CEPH and i should not have issues to connecting as additional storage.
I think here i can store VM, but only in raw format, this is right?

Option B)

Additional Disks

Classic Volume - 200 IOPS guaranteed - 0.04 €/per GB/month
High Speed Volume - Up to 3,000 IOPS - 0.08 €/per GB/month

With this kind of storage - built on Openstack- is of course "pay to use" it seems i have to pay monthly for every instance + the GB spent.


In my case, as i previously commented i need a fast storage (i have local SSD), so the High Speed Volume (the most expensive) will be the best for me.

I don´t know if Object Storage is a good option for me.

So, it will be nice hear all kind of comments about how you manage this attached storage, and if anybody is using this OVH storaging please comment about aproximately is possible the final price: a standard dedicated server with their physical disks have a fixed cost so any orientation in this sense will be high appreciated.

Thanks to all!

 

[amil]

Any suggestion, please?

 

[spirit]

you can't store vm disk on object storage, you need a block storage.

ceph have multiple layers, objects (rados) , this is like amazons3 , or rbd (rados block device) for vm disk.

I think the "additional disk" offer is rbd block devices

 

[amil]

Thank you spirit for your reply: so i understand that "additional block devices" it will valid for me and i supposed i could connect this remote space to proxmox, right?

They said for "additional disk": These disks can be configured from 10 GB to 10 TB and are available in two versions: one has the performance of a hard drive, and the other can be compared to an SSD drive.

Did you work with rbd block devices on your servers spirit?


Thank you so much

 

[spirit]

I don't known how they expose the block device, but it "could" be possible to add rbd storage in proxmox (add ovh ceph monitors ips and ceph key). then manually add rbd volume in vmid.conf the disk.

but you should ask to ovh to be sure.
(maybe it's only available for openstack, because openstack manage the disk creation).

I'm using ceph myself, but I manage my own ceph storage in my datacenter. (with proxmox rbd plugin to create disk)

 

[amil]

Nice to hear spirit, in proxmox web interface i also can see also "RBD" as option adding disk to pool, but as you suggested i am going to contact now with ovh to ensure how their work and if i am able to connect this space to my proxmox.

I´ll update soon with news

 

[amil]

Finally, i got the ovh response, and no good news.... i paste here literally -translated to english- the e-mail response:

"""
Dear customer,

I can confirm that is NOT using RDB.

If you use ovh Public Cloud to host VM machines is NOT recommended absolutely, you will have heavy connectivity problems.

"""

So, they not give me more information about their tech, only the afirmation that is a very bad idea, so very bad news for me.

160 GB SSD in real storage -in every dedicated server- is very few in terms of local disk space...

I don´t know if another "cloud storage" will work as expected (ie. AWS or Google)...

Of course build now, our own RDB storage is not possible due its costs, so we are try to find some "middle" solution.


In other order of things, the 3x SSD hd on raid 5 gives me amazing results with an hdparm:

root@host:~# hdparm -Tt /dev/mapper/pve-data

/dev/mapper/pve-data:
Timing cached reads:  28180 MB in  2.00 seconds = 14106.24 MB/sec
Timing buffered disk reads: 2152 MB in  3.00 seconds = 716.85 MB/sec
root@host:~# hdparm -Tt --direct /dev/mapper/pve-data

/dev/mapper/pve-data:
Timing O_DIRECT cached reads:  872 MB in  2.00 seconds = 435.08 MB/sec
Timing O_DIRECT disk reads: 1356 MB in  3.00 seconds = 451.52 MB/sec

Really, i have multiple doubts of the performance of shared / RBD storage.

Any point of light will be high interesting.

For all those using modest hardware with SSD hd, what is your workaround?