Understanding DB2 HADR, HACMP, SQL Replication, and Storage DR for High Availability

1. DB2 HADR

HADR (High Availability Disaster Recovery) is IBM's database‑level data‑replication feature originally derived from Informix HDR. It provides a primary‑standby pair; before DB2 9.7 the standby was not readable, while from 9.7 onward it can be read to offload the primary. Synchronous transmission is recommended when bandwidth is ample and zero data loss is required; asynchronous or super‑asynchronous modes are used for long‑distance links (e.g., Beijing‑Shanghai) where latency or bandwidth is limited. HADR cannot be used with DPF, works only for single‑partition databases, and does not support data compression, encryption, or heterogeneous replication. Compared with Oracle DataGuard, HADR offers faster failover with fewer failures, but lacks built‑in compression/SSH integration.

2. SQL Replication and Q Replication

SQL replication is suited for same‑LAN environments, while Q replication (leveraging WebSphere MQ) tolerates poorer networks by buffering changes. Q replication is often combined with HADR for remote disaster‑recovery (e.g., China Tobacco’s DR site) and works by analyzing transaction logs, imposing minimal performance impact. It supports table‑level replication and is well‑integrated with DB2; support for Oracle and other databases is limited. IBM’s CDC (formerly Data Mirror) and Oracle GoldenGate provide similar capabilities, though CDC can be complex when handling table dependencies.

3. HACMP

HACMP offers three clustering modes:

Cascading : Primary and standby nodes have priority; resources run on the highest‑priority node. After a failure, resources move back to the primary.

Rotating : Nodes share equal priority; resources start on the first node and remain there after failover.

Concurrent : No primary/standby distinction; resources run on all nodes simultaneously, providing continuous availability without resource migration.

Rotating mode fits high‑availability telecom services, while Concurrent mode pairs well with Oracle RAC or parallel servers. HACMP alone does not replicate database data; it must be combined with HADR for full protection.

4. DPF High‑Availability Considerations

DPF itself lacks built‑in HA mechanisms, but multi‑node configurations can provide limited disaster tolerance if critical catalog nodes remain up. Table‑level availability depends on the partition nodes hosting the tables; if a node fails, only tables on that node become inaccessible. OS or network failures can be mitigated with HACMP, but database corruption on a node cannot be recovered automatically. Proper planning of HA for catalog nodes and regular verification of table‑space distribution are essential.

5. Storage‑Level Disaster Recovery

Storage‑level DR complements database HA. Technologies such as EMC SRDF provide synchronous, near‑synchronous, and asynchronous replication across distances ranging from a few kilometers to several thousand kilometers, supporting virtually any host or database platform. For short distances, direct fiber is preferred; for longer spans, DWDM solutions from vendors like Huawei or Cisco are used. When budget constraints exist, tape‑based migration or less expensive replication may be employed for less critical data. Veritas BMR offers block‑level backup and recovery without requiring LAN SAN bandwidth, with agents for various databases to enable hot backups.

6. Network, Power, and Organizational Measures

Network redundancy (multiple NICs, subnets, redundant fibers, and at least four switches and storage hosts between primary and DR sites) is mandatory. Power redundancy requires UPS units and generators. Large telecom operators and banks follow strict operational procedures and regulatory guidelines (e.g., People’s Bank of China) to control privileged access, prevent human error, and conduct regular DR drills at both infrastructure and application levels.

The author acknowledges that the discussion is incomplete and invites further community input to refine the understanding of enterprise‑grade database high availability.