SANs are important data center components for mass storage and enterprise data management, but in MPP environments they are limited by their finite channel throughput. Conversely, an MPP server’s direct-attached storage (DAS) is fast and scalable, but is an unmanaged island of storage requiring separate backup and disaster recovery support. PADB’s patent-pending Blended Scan interleaves SAN I/O with DAS I/O to overcome these disadvantages and make better use of the configuration’s entire DAS- and SAN-based storage capacity.
PADB can integrate the two storage environments (DAS and SAN) using the SAN for the server’s RAID 1 pair and for a non-equal portion of the total I/O processing. This way PADB can benefit from the DAS I/O throughput and take full advantage of the SAN I/O throughput, too. Plus, it can now leverage the enterprise data management features of SANs.
SAN-Based Node Failover with HSN
PADB’s enhancements for SAN environments (see “High Performance in SAN-Based Environments”) allow it to leverage a SAN-specific approach for high availability (failover/failback) if server hardware fails. In this scenario, all storage is mirrored on the SAN so the hot spare can immediately take over processing by accessing the SAN-based copy of the data.
Similar to the non-SAN environment, each node’s internal disks are mirrored to multiple logical units (LUNs) in the SAN. However, the SAN’s LUNs are always configured with RAID so disk spares or RAID are not needed for the server’s internal drives.
For the purposes of snapshot backups and disaster recovery, the mirror images on the SAN are considered the “data of record”. Thus, the server’s internal disks do not directly participate in the HA model, but instead function as a “cache” for part of the SAN.
PADB stores temporary tables and other transient data on SAN partitions that do not participate in the snapshot or Disaster/Recovery (D/R) system. This significantly reduces the overhead of snapshot maintenance and reduces D/R connectivity requirements.
Disaster Recovery (D/R)
For higher assurance of continued processing, disaster recovery systems can be implemented locally or remotely by leveraging a SAN’s enterprise software features (e.g., EMC’s Navisphere Management Suite). In this scenario, the enterprise software handles data replication and synchronization.
Snapshots (Snaps)
Snaps instantly create and maintain a record of the database at a point in time to provide for recoverability from processing errors. Snaps provide a low-overhead way for customers to restore the database system to a prior state in a matter of seconds, rather than experience a time-consuming restore from backups.
PADB leverages the SAN’s enterprise management software (e.g., Navisphere) to initiate and manage snaps. Snaps can be scheduled to run periodically, or invoked dynamically by the management software or a SQL instruction to PADB (which in turn notifies the management software).
PADB’s ability to initiate snaps is a unique advantage. Managing the snap transaction from within the database ensures the snap data is at the same transaction commit level even across multiple SAN and D/R instances without acquiescing the database. This allows processing to continue while also eliminating the possibility of D/R systems being out of synch.