VAST Cluster listens for requests on IP endpoints that are virtual, and therefore called Virtual IPs (VIPs). These endpoints are for all data traffic for all protocols (currently NFS and S3).
To configure the VIPs, you need to create VIP pools.
VIP pools are ranges of IP addresses that VAST Cluster can use to listen for data traffic. All VIPs in all configured VIP pools are distributed evenly among all active CNodes*. If a CNode fails, the VIPs assigned to it are automatically moved to other active CNodes, ensuring that clients can continue to connect to stable IP addresses.
This model provides for load balancing and transparent failover. Load balancing is achieved by configuring a 'round robin' load balancing configuration on your DNS server. All of the VIPs in all configured VIP pools should be mapped to a DNS name that you configure on your client-side DNS server, in a 'round robin' load balancing configuration.
*unless you choose to limit a VIP pool to a specific group of CNodes. See Dedicating CNode Groups to Specific VIP Pools.
VAST Data recommends a minimum of two VIPs per CNode. For optimal load balancing, we encourage four VIPs per CNode for clusters with one Cbox (four CNodes); and four or more VIPs per CNode for larger clusters.
When determining how many VIPs to configure, consider the following:
Not only should there be at least one VIP available to each CNode, but since each CNode has two Network Interface Cards (NICs), there should be at least two VIPs available to each CNode.
It is desirable for the number of VIPs to be evenly divisible by the number of CNodes. That way when all CNodes are running, VIPs will be spread evenly.
More VIPs improve balancing on failure. In case a CNode with only one VIP were to fail, the one VIP for that node would be moved to one active CNode, doubling the work of that CNode. If the failed CNode had more than one VIP, each VIP could be moved to a different active CNode. Therefore, the more VIPs there are per CNode, the better the system is able to balance load on failure. With one VIP per CNode, 100% of that load is moved to one other CNode. With two VIPs, 50% is moved to one and 50% to another. With three VIPs, the ratio is 33%/33%/33%, and so on.
- From VAST Cluster 2.0.8, if the cluster has four or more CBoxes, the management CNode (the CNode that runs VMS) is not assigned VIPs. Therefore, for clusters with four or more CBoxes, in order to evenly assign four VIPs per CNode, assign four times one less than the total number of CNodes.
CNode groups are supported from VAST Cluster 1.2.3.
You can limit a VIP pool to a specific group of CNodes, in order to dedicate those CNodes to a specific set of hosts or applications.