What do you think of my hypothetical HCI
-
@dashrender said in What do you think of my hypothetical HCI:
I see storage here
I assume RAID 10 is enough reliability?
For SSDs you could get away with RAID 5. Would free up more space for you.
-
@dashrender said in What do you think of my hypothetical HCI:
I see storage here
Ya I see it in the drawing... but where does it live. It shows 3 KVM compute-only nodes. I took that to mean no local storage for VMs. So back to my question, where does the storage live? On a single storage host?
-
@tim_g said in What do you think of my hypothetical HCI:
@dashrender said in What do you think of my hypothetical HCI:
I see storage hereYa I see it in the drawing... but where does it live. It shows 3 KVM compute-only nodes. I took that to mean no local storage for VMs. So back to my question, where does the storage live? On a single storage host?
Oh I see now, two separate storage hosts, using gluster.
-
Pure computer + external storage is the opposite of HC. It's not converged at all here, this is an IPOD design.
I'm still unclear where the storage is. I see pure compute nodes (why do you want pure computer?) and I see storage nodes for backups, but I don't see any obvious storage for the primary running VMs.
-
@scottalanmiller said in What do you think of my hypothetical HCI:
Pure computer + external storage is the opposite of HC. It's not converged at all here, this is an IPOD design.
I'm still unclear where the storage is. I see pure compute nodes (why do you want pure computer?) and I see storage nodes for backups, but I don't see any obvious storage for the primary running VMs.
Top right corner. He wants to use a GLUSTER setup to run the VMs from.
-
To make things cheaper, I know its separate (which is not HCI) but the whole point is to allow scale in and scale out which the main purpose of HCI (I think), maybe it is not HCI but its software defined alot, right ?
And Storage for servers is expensive, and why do you want them if you can make KVM server just state, if one goes does you have another compute node that can mount the storage and proceed, regarding automation of that task perhaps the saltmaster node can do script for this.
-
@tim_g
Hi,
The storage is on seperate storage nodes, RAID 10 nodes and glusterfs + distributed mode to tie them together with other storage nodes, also backup nodes exists while role is to take snapshots of VM qcow2 images each week/month.
-
again sorry for the poor drawing, it is small board I have in my room but the idea of it is to allow scaling out in easy manner, which is what I think is the main purpose of HCI as I perceived it, also it is meant to be as free as possible software wise and software defined (I hope). Also I reply late due to GMT +3
-
@emad-r said in What do you think of my hypothetical HCI:
The storage is on seperate storage nodes,
That would make it an IPOD and the opposite of HC. It's diverged, not converged. What's the purpose of so many nodes, you have five where only two or maybe three are needed.
Also needs to be noted, Gluster needs three nodes minimum.
-
@emad-r said in What do you think of my hypothetical HCI:
again sorry for the poor drawing, it is small board I have in my room but the idea of it is to allow scaling out in easy manner, which is what I think is the main purpose of HCI as I perceived it,
No, that is not the purpose of HC. However, HC allows for easier scaling simply by being a more logical design. Diverged (IPOD) like this requires more nodes for everything and makes scaling slightly more difficult, rather than simpler.
-
@emad-r said in What do you think of my hypothetical HCI:
RAID 10 nodes and glusterfs + distributed mode to tie them together with other storage nodes, also backup nodes exists while role is to take snapshots of VM qcow2 images each week/month.RAID 10 is not normally used with RAIN systems like Gluster. You'll end up with really bad utilization rates.
-
I see , the whole main concept of this is tie 2 RAID system together or basically scale RAID :
Distributed - Distributed volumes distribute files across the bricks in the volume. You can use distributed volumes where the requirement is to scale storage and the redundancy is either not important or is provided by other hardware/software layers.
So I will work on it more to merge everything together in one node but having the traditional RAID makes me feel more comfortable than going all RAIN. But i will look into improving it, by making the CPU servers with storage and meshing everything together.
-
@emad-r said in What do you think of my hypothetical HCI:
I see , the whole main concept of this is tie 2 RAID system together or basically scale RAID :
That's just basic storage, you should be doing that with any type of storage. That's how SAN, NAS, everything has to make high availability. That's not related to *vergence.
To scale RAID, you need Network RAID, that's DRBD, HAST, Starwind, etc.
Or the alternative is RAIN, like Gluster, CEPH or Scale HC3's SCRIBE.
-
@emad-r said in What do you think of my hypothetical HCI:
So I will work on it more to merge everything together in one node but having the traditional RAID makes me feel more comfortable than going all RAIN.
That's an emotional reaction. Both have their place, but you need to focus on the goals, and take the emotions out of the situation. If you don't trust RAIN, you can't use it period. If you don't trust it and use it anyway, that creates a problematic situation.
-
Why not just use two or three Hyper-V Server 2016 nodes and StarWind vSAN?
That seems like a whole lot easier, cheaper, and way more effective.
-
@tim_g said in What do you think of my hypothetical HCI:
Why not just use two or three Hyper-V Server 2016 nodes and StarWind vSAN?
That seems like a whole lot easier, cheaper, and way more effective.
And you totally missed "massively safer." Better storage, and way fewer moving pieces to fail.
-
@scottalanmiller said in What do you think of my hypothetical HCI:
@tim_g said in What do you think of my hypothetical HCI:
Why not just use two or three Hyper-V Server 2016 nodes and StarWind vSAN?
That seems like a whole lot easier, cheaper, and way more effective.
And you totally missed "massively safer." Better storage, and way fewer moving pieces to fail.
Yeah that too of course! That was bundled into the "way more effective" bit.
