cancel
Showing results for 
Search instead for 
Did you mean: 

I/O fencing disk for SFRAC for Extended RAC

Xentar
Level 4

Hi All,

We would like to have an extended RAC setting with SFRAC.

The environment as the following, there is 2 hosts, each connected with 2 shared storages.

We have an question about how the VCS I/O Fencing disk to be assigned?

3 LUNs from each storage and map to each host? And then add those 6 disks in to vxcoorddg?

Regards,

Xentar

1 ACCEPTED SOLUTION

Accepted Solutions

mikebounds
Level 6
Partner Accredited

You must have an odd number of fencing objects with a minimum of 3.  So with only 2 arrays this would mean having 2 LUNs from one array and one from the other array.  This is far from ideal as if your arrays are on seperate sites and you loose site with array containing 2 disks, then the other node will race to get disks, only be able to get its one local fencing disk and hence VCS will panic node as the node must get 2 or more LUNs to survive.  If you don't have a 3rd array, then I would recommend using CPS, where you use 2 I/O fencing disks and one CPS.  CPS stands for Coordination Point Server, so that it in effect acts as an I/O fencing disk which is raced for in split-brain or loss of node scenarios.  CPS has to be in a VCS cluster, but if doesn't need to be dedicated or the same platform so you can put CPS on any other cluster you have.  If you don't have any other clusters, you can setup a one node Linux or Windows cluster which can be in a VMWare partition

Mike

View solution in original post

2 REPLIES 2

mikebounds
Level 6
Partner Accredited

You must have an odd number of fencing objects with a minimum of 3.  So with only 2 arrays this would mean having 2 LUNs from one array and one from the other array.  This is far from ideal as if your arrays are on seperate sites and you loose site with array containing 2 disks, then the other node will race to get disks, only be able to get its one local fencing disk and hence VCS will panic node as the node must get 2 or more LUNs to survive.  If you don't have a 3rd array, then I would recommend using CPS, where you use 2 I/O fencing disks and one CPS.  CPS stands for Coordination Point Server, so that it in effect acts as an I/O fencing disk which is raced for in split-brain or loss of node scenarios.  CPS has to be in a VCS cluster, but if doesn't need to be dedicated or the same platform so you can put CPS on any other cluster you have.  If you don't have any other clusters, you can setup a one node Linux or Windows cluster which can be in a VMWare partition

Mike

Gaurav_S
Moderator
Moderator
   VIP    Certified

I agree with Mike, using CPS would be a better choice here in comparision to spliting coordinator disks...

 

G