Hardware

[iruindegi]

Hi,

We recently buy a new server (https://www.supermicro.com/products/chassis/2U/825/SC825TQ-R720LP.cfm) with 2 CPU, 3 SSD 500GB and 32 Ram memory. Installed proxmox and everything is working fine (now we are going to buy the subscription).

We want a setup with HA, and my question is about harware. We are looking foward to buy 2 HP Microserver gen 8 ( (2,3 GHz, Intel Xeon, E3-1220LV2, 8GB Ram) and some disk to complete a 3 server cluster. Our VM machines will be launched on Supermicro's server and one of the microserver for backup tasks.

Is that ok? Do we need better hardware os is it ok with this? Firstly we are thinking about Ceph but we can go with nfs too.

Any observation or advise will be appreciated.

Thanks in advance

 

[LnxBil]

Ceph with 3 nodes is possible, but not recommended. Please do not buy consumer SSDs, you will not have fun with them.

Getting the shared storage for you system will be the hard part: NFS is not HA, and can therefore not be used directly. You have to use Ceph or glusterfs. I'd give both a test spin and see how well they perform.

 

[iruindegi]

Ok. So will go with Ceph.

Yeah we always buy server oriented drives. Our idea is to buy the Hp Microserver with a ssd (only for the system) and some HDD for the cluster data storage.

What I want to know is if the hardware is ok

 

[spirit]

Ceph with 3 nodes is possible, but not recommended.

? why ???

I have a lot of small 3 ceph nodes clusters (18 osd by cluster), with full ssd , getting 600k iops read 4k , 150k iops write 4k.
That's works really good.

 

[LnxBil]

? why ???

I have a lot of small 3 ceph nodes clusters (18 osd by cluster), with full ssd , getting 600k iops read 4k , 150k iops write 4k.
That's works really good.

I'd say speed, but your machines are quite beefy and disproof my point :-D

Could you describe your machines a bit better? What SSDs are you using? 2 or 3 copies of data?

 

[debi@n]

Getting the shared storage for you system will be the hard part: NFS is not HA, and can therefore not be used directly. You have to use Ceph or glusterfs. I'd give both a test spin and see how well they perform.

sorry? we are working with NFS and HA proxmox. nfs with synology HA. what´s the problem?

 

[gosha]

sorry? we are working with NFS and HA proxmox. nfs with synology HA. what´s the problem?

Maybe it is meant that the NSF is not clustered storage and can itself become a point of failure.
(If itself not placed in other HA cluster of course )

Best regards,
Gosha

 

[spirit]

I'd say speed, but your machines are quite beefy and disproof my point :-D

Could you describe your machines a bit better? What SSDs are you using? 2 or 3 copies of data?

replication x3 , each node : 2 x xeon x10 cores 3,1ghz , 64g ram , 6 intel s3610 1,6TB.

The more iops you need, to more cpu you need. (But I can reach easily 10GB with small cpu and big blocks size)

 

[iruindegi]

so... hp microservers will be enough?

 

[LnxBil]

Maybe it is meant that the NSF is not clustered storage and can itself become a point of failure.
(If itself not placed in other HA cluster of course )

Exactly what I meant. There should not be a single point of failure inside your data center for real HA. Anything can and will fail.

 

[LnxBil]

replication x3 , each node : 2 x xeon x10 cores 3,1ghz , 64g ram , 6 intel s3610 1,6TB.

The more iops you need, to more cpu you need. (But I can reach easily 10GB with small cpu and big blocks size)

Thanks.