AnsweredAssumed Answered

Suggestions for Cluster Hardware Design and Services Node Layout

Question asked by dzndrx on Mar 28, 2017
Latest reply on Apr 14, 2017 by aalvarez

Hi Im currently working on a 3 node mapr cluster, Here are the details below.

 

Use case: MapR Cluster as NFS data storage for archiving and Virtual Machine datastore (ESXI).

 

Objectives:

  • MapR to serve as a data lake, dumping all data to MapR Cluster through NFS.
  • MapR-FS as a datastore to ESXI through NFS.
  • MapR-FS to be mount on Virtual Machines for application storage through NFS.

 

Demo will be successful if they find noticeable speed increase on objectives. 

 

Environment: MapR 5.2 

Node Specifications: 3 nodes. Per node (12 CPU Cores + 32 GB Memory + (1 TB x 6) + 2 NICs @ 10GBe each)

Client Internal Network Speed: 10 Gbe

Node layout (default)

 

So what we want to do is to improve the existing data center by providing MapR cluster as a storage to leverage in its NFS capabilities, also aiming to future proof for business analytics since all the data resides on the MapR cluster.

 

The node specification can be upgrade on clients request given they find an improvement on the objectives above. Currently, we don't have the exact number (time) on their archiving. But they provide some details about their applications that run in the Virtual Machines. I/O extensive applications.

 

Generally my plan is to have a single gateway for NFS and to use a VIP for HA on setting up the NFS. For example:

 

VIP: 192.168.0.1

Node1 dedicated to: 192.168.0.1

 

Datastore on ESXi

Datastore1 mounted on 192.168.0.1

*All VM's are now using Datastore1 and inside each VM's a folder mounted on 192.168.0.1 through NFS.

 

Incase node1 is down node2 or node3 automatically kick in as the new gateway.

 

On the other hand I can also set a 3 VIP having each node as a gateway to serve more clients or to provide some load balance by setting up 3 datastores in the ESXi each using 3 different NFS gateways. For example:

 

VIP: 192.168.0.1 ~ 192.168.0.3

Node1 dedicated to : 192.168.0.1

Node2 dedicated to : 192.168.0.2

Node3 dedicated to : 192.168.0.3

 

Datastore on ESXi

Datastore1 mounted on 192.168.0.1

Datastore2 mounted on 192.168.0.2

Datastore3 mounted on 192.168.0.3

 

*For exampe we have 9 VM's, each Datastore will handle 3 Vm's giving the equal load to each gateways. And if a node is suddenly down VIP will kick in.

 

What would be the best design? Any suggestions from the community?

Outcomes