ONTAP Discussions

cDOT really a scale out solution?

Matt99
3,145 Views

Hello Everyone,

 

We are a long time netapp partner trying to get a handle on cDOT marketing language.

 

I've gone through several cDOT trainings and read several guides on cDOT but I'm simply not seeing how cDOT is a scale out platform?  We have a customer that wants to move to 8.3 for this reason.  But after engaging with Netapp for anwsers the picture is even less clear. 

 

From what we have can figure out (via netapp support and lab setup).  cDOT clusters do not span aggragetes, raidgp across clusters.  Volumes can but use somthing like a junction point.

 

So here is an example,

Customer wants a new cluster/scale out NFS volume for VMWare ESXi.   That single NFS datastore will reside on an aggragate that is fully contained within a single node.  If I add another node to the cluster, the datastore data still stays on that single node.   I can move the volume around for sure, but the data is served from that single node.  Typically in scale out, the data is re-stripped across nodes and hence scales out across the new spindles and cpu/ram of more nodes.

 

In a VMWare setup the NFS path of  CLUSTER1:/vol/NFSVOL1 will not write to mulitple nodes or read from multiple nodes.   Sure I could do CLUSTER1:/vol/NFSVOL1 and CLUSTER1:/vol/NFSVOL2 which can come from different nodes, but I could do that with 7-mode except the node name is not hidden behind the cluster.

 

How is that scale-out?  Preformance doesnt go up by adding nodes.  I get a unified name space, but thats not a preformance related feature.....just a management thing.

 

I know in general 8.3 has alot of other nice things....but this scale-out term is really confusing our customers or I just dont undersand how cDOT works.   

 

Do I have this wrong?

 

 

 

 

3 REPLIES 3

aborzenkov
3,143 Views

Did you consider infinite volumes?

Matt99
3,138 Views

I have.  If I'm not mistaken they work as I have described above.

 

Maybe we tested them wrong but here is how it worked for us

 

VOL1

      Subdir 1 -------> Node A

      Subdir 2 -------> Node B

 

Like a junction point.  But thats not scale out as preformance for any one particular dataset does not increase with adding nodes.  If there is a critical Database file under Subdir2, the only way to increase its preformance is scale-up (ie more disk to that node, flash etc). 

 

Its even more complicated with ESXi as it writes all things to the root of a volume.

 

-M

aborzenkov
3,132 Views

Yes, each file is located inside of a single constituent, that's right. So it will scale with number of VMs (or better vmdks) but not for single vmdk.

Public