After a short break, I would like to continue SAP HANA High Availability and Disaster Recovery article series with a high level overview between the two most used high availability options in SAP HANA: System Replication and Host Auto-Failover.
Host Auto-Failover is basically an automated failover from crashed host to a standby host in the same system. In this method, SAP HANA regularly checks if all the cluster members are still active; and when an active host fails, a standby host “automatically” takes over its place. An internal cluster manager “nameserver” manages this entire failover process. Host Auto-Failover is a very reliable HA solution and provides absolute zero data loss.
System Replication is the continuous update of secondary system(s) by primary system, combined with different replication and operation modes which also provides in-memory table loading. It is a standard SAP HANA feature and can be used as an HA or DR solution. Because all data is replicated to the secondary site and data is pre-loaded into memory, the recovery time objective is reduced significantly.
The previous articles about Host Auto-Failover and System Replication contain useful detailed information you need to know about both and I believe below table would be more suitable to provide a high-level overview between two methods.
Syncmem: Primary node waits until data is received by secondarySync: Primary node waits until data is received and persisted by secondaryFull sync: transaction processing on the primary node is blocked until secondary system becomes availableAsync: primary node does not wait any acknowledgement or confirmation from the secondary node
|Host Auto-Failover||System Replication|
|Description||Cluster-like solution in the same data center||Similar to classical shadow database solution|
|Can be used for||High Availability||High Availability and Disaster Recovery|
|RPO||Zero||Zero with synchronous replication|
|Architecture||“N+M” HA/fault recovery solution. One or more Standby nodes, data is not preloaded.||N+N HA/DR solution. Secondary has the same number of active nodes, data is pre-loaded.|
|Automated fail-over||Available on the host level (the failure of a single service does not trigger the failover). Failover is internally managed by SAP HANA nameserver.||Available with external cluster management software provided by the hardware vendors.|
|Data pool||One data pool, shared NFS or distributed filesystem.||Multiple data pools, single or multiple data centers support.|
|Data consistency||Heartbeat (regular TCP communication) and I/O (isolating the failed node) fencing options||Continuous replication with different replication and operation modes|
|Multitenant Database Container Support||Available on the host level||Replication of the whole system, single tenant replication is not possible.|
|Replication||Not required as disk is shared and data consistency is already achieved||Syncmem: Primary node waits until data is received by secondary
Sync: Primary node waits until data is received and persisted by secondary
Full sync: transaction processing on the primary node is blocked until secondary system becomes available
Async: primary node does not wait any acknowledgement or confirmation from the secondary node
|Scalability||Better suited for scale-out||Better suited for scale-up|
|Flexibility||Rigid||Secondary node can be used for QA with additional disk|
As you can see, both methods are quite different from architecture to failover process. I believe the best approach for HA/DR solution should start with determining business requirements properly, followed by the effective sizing of SAP HANA and purchasing the correct hardware if you are implementing on premise. During the sizing phase, you should decide which high availability, fault recovery and disaster recovery methods would meet the business requirements without breaking the bank. Detailed analysis of each method would significantly reduce the risk of wrong architecture.
Have any questions about SAP HANA System Replication or Host Auto-Failover?
Leave a comment below.
If you liked this post, you might like these relevant posts: