cancel
Showing results for 
Search instead for 
Did you mean: 

Can we us heartbeat NIC to connect with a IP device ?

Zahid_Haseeb
Level 6
Partner    VIP    Accredited

Suppose a 2 nodes VCS HA environment have 2 local heartbeats. As per my understanding VCS dont need IP addresses for its local heartbeats communication and uses its own protocol(LLT) for heartbeats between both HA nodes.
My question is, if we assign an IP address on one heartbeat NIC and connect a ip device back to back, Can we allow to do it ?

6 REPLIES 6

frankgfan
Level 6
   VIP   

Hello,

Technically yes you can do it although it is not recommended for a failover VCS configuration.

As Veritas cluster requires minimum two dedicated NICs for cluster inter node heartbeat (primarily) communicaton on dedicated private network meaning if an IP is plumbed on a private NIC, only the nodes in the cluster can use this IP for sending/receiving network packets.  For testing purpose, its perfectly OK to plumb an IP on one or both NICs.

However if you think of transmiting large amount of data through these NICs via LLT private network, it is not recommended.  When the large amount of packets is transmitted through the LLT network, it may results cluster nodes miss heartbeat and errors like node inactive/in trouble may be resulted.

For CVM/CFS/RAC clusters, they are different and beyond the scope of this discussion.

Regards,

Frank

Zahid_Haseeb
Level 6
Partner    VIP    Accredited

Actually we have four NICs and all are occupied. Any recommandation if we can go with a low priority heartbeat with single LLT heartbeat ?

The absolute minimum is one private and one (low priority) public.

But, I would not bet my paycheck on such a configuration. As others have said, two private links, using different NIC cards and switch infrastructure, along with one or more low-propriety (public) links, are what is recommended for production. The cost of NICs and switch ports are very small compared to time spent troubleshooting an unstable cluster and associated downtime. The above recommendation is for ANY cluster technology, not just VCS.

I suggest reading the Admin Guide and refer to the sections that mention failure scenarios and how the cluster behaves when down to one communications path (jeopardy).

Cheers

_______________________________
My customers spend the weekends at home with family, not in the datacenter.

Zahid_Haseeb
Level 6
Partner    VIP    Accredited

Agreed that 2 LLT heartbeats and one low priority heartbeat is an ideal scenario. Additionally, I can understand that price for NIC & switch is small but actually the hardware are blade servers and no provision to add additional NIC in it.

 

To ensure reliability of cluster on a one LLT heartbeat and one low priority heartbeat, can we configure SAN BASED FENCING on top of one LLT heartbeat and one low priority heartbeat cluster ?

Fencing is separate, but related, to communications. Some type of fencing is always highly recommended for all clusters in production. You have the choice of disk or CPS. Three or more disks (odd number) or three or more CPS (odd number). Disks must support SCSI -3 PGR.

LLT is for cluster communications. Cluster fencing guarantees cluster membership when members are not responsive (network failures, over-burdened links, memory/cpu starved OS, unresponsive OS, etc). Implement priority-based fencing feature when possible to make sure the best node wins. See Admin Guide for details.

Regarding your NICs: modern blade systems have lots of bandwidth, usually with 10Gb links. Single blades can rarely saturate even one 10Gb link. Why not redistribute the links? Just an idea.

Cheers




_______________________________
My customers spend the weekends at home with family, not in the datacenter.

Zahid_Haseeb
Level 6
Partner    VIP    Accredited

Why not redistribute the links?

Please elaborate ?