Common Site Interlink and Witness Requirements
The following describe general requirements, and the HXDP release they apply to.
Network Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
10 Gbps dedicated, 5-ms Round-Trip Time (RTT) latency between the two active sites is required. |
✔ |
✔ |
100 Mbps, 200 ms-RTT worst case latency for 16 kilobyte packet sizes between the active sites and witness site is required. |
- |
✔ |
Existing fabric interconnects are supported, provided the fabric interconnects support M5 or M6 servers. |
✔ |
✔ |
User VMs should be capable of vMotioning to any site, without impacting external network connectivity to these VMs. |
✔ |
✔ |
Similar to regular HX Clusters, two separate IP subnets are required—Both over Stretch L2. One subnet for data traffic and one for management traffic, with the management subnet reachable from vCenter and the witness node. |
✔ |
✔ |
FI facing ports need to have Port-fast, spanning-tree port type edge trunk, or similar spanning tree configuration that immediately put ports into forwarding mode. |
✔ |
✔ |
QoS If the HyperFlex cluster is a stretch cluster, you should enable QoS end-to-end between the sites. The QoS policies in the Cisco UCS and upstream network should also be aligned so that HyperFlex traffic can receive consistent QoS end-to-end. If the HyperFlex cluster is a stretch cluster that connects to an ACI Multi-Pod fabric, you should enable QoS policies in the ACI fabric and extend it across the IPN as well. The QoS policies in the Cisco UCS and ACI fabrics should also be aligned so that HyperFlex traffic can receive consistent QoS end-to-end. There may be a potential conflict between the Quality of Service (QoS) policy for HyperFlex and Nexus 1000v. Make sure that the QoS classes for N1Kv are set as per the HyperFlex policy. See Creating a QoS Policy, in the Network and Storage Management Guide. |
✔ |
✔ |
Network Topology
![](/c/dam/en/us/td/i/300001-400000/300001-310000/306001-307000/306439.jpg)
Witness VM Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
Configure and enable NTP on all servers. |
- |
✔ |
An independent third witness site is required. |
- |
✔ |
A separate witness is required for each cluster. |
- |
✔ |
Both the main sites must have connectivity to the third witness site with a minimum bandwidth of 100 MBPS, 200 ms-RTT worst case latency for 16 kilobyte packet sizes. |
- |
✔ |
Site must have the capability to deploy and run Open Virtualization Format (OVF) image. |
- |
✔ |
Network Latency Requirements for the Witness VM
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
The HyperFlex Stretch Cluster solution requires that the Witness VM be located in a third site to ensure that a storage site failure does not affect the Witness VM. |
- |
✔ |
The solution can support a Witness bandwidth as low as 100 Mbps, 200 ms-RTT worst case latency for 16 kilobyte packet sizes. |
- |
✔ |
Latency to the witness impacts site failure times, and it is recommended that for larger clusters with significant load and data, to have RTT times in the order of 10ms or lower. |
- |
✔ |
Witness VM Fabric Interconnect and Node Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
Symmetric configuration is required across both sites. |
✔ |
✔ |
There must be a minimum of two converged nodes on each site. |
✔ |
✔ |
A maximum of 16 converged nodes on each site is supported. Ensure that both sites have the same number of converged nodes. |
✔ |
✔ |
There must be a redundant fabric interconnect configuration on each site. |
✔ |
✔ |
Converged nodes have to be M5 or M6 nodes. |
✔ |
✔ |
Ensure that the Fabric Interconnect pair is of the same model in the same domain. |
✔ |
✔ |
Compute-only nodes are supported. |
✔ |
✔ |
VMware Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
VMware Enterprise Plus edition with HyperFlex Stretch Cluster is highly recommended to ensure proper failover behavior and guarantee high performance during normal operations. While it is possible to run Stretch Cluster without VMware Enterprise Plus edition, the advanced DRS features such as site affinity will not be available, negating some of the intended operational functions of Stretch Cluster. |
✔ |
✔ |
Use a single vCenter for both sites. |
✔ |
✔ |
The vCenter can be a VM running at the same site as the witness. |
- |
✔ |
Nested vCenter is not supported in Stretch Cluster. |
✔ |
✔ |
The vCenter must be configured independently for High Availability, as required. |
✔ |
✔ |
UCS Manager Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
Two separate, manually synchronized fabric interconnect domains are required. |
✔ |
✔ |
VLAN Requirements
Requirement |
HXDP Release 5.5(1a) and Later (Invisible Cloud Witness) |
HXDP Release 5.0(x) and Earlier (Witness VM) |
---|---|---|
IP addresses for nodes on both sites are required. |
✔ |
✔ |
Stretch VLANs across both sites are required. |
✔ |
✔ |