6.3 B10000 Resiliency & Data Services: All-Active Nodes, RAID 6, Data Reduction & Replication

B10000 Resiliency & Data Services

Quick Answer: The B10000 stays available because all controller nodes are active, there is no single point of failure, drives are protected with RAID 6 and distributed sparing, and software is upgraded non-disruptively. Its data services include inline data reduction, Virtual Copy snapshots (including immutable ones), Remote Copy replication and Active Peer Persistence metro clustering, which with an HPE Quorum Witness fails over automatically.

All-active, no single point of failure

• All nodes serve I/O. There is no idle standby controller. Volumes are spread across nodes and drives (wide striping), so the whole system shares the load.
• Shared access to capacity. In switched systems every controller node sees every JBOF. If a node fails, the surviving nodes keep serving its volumes and no capacity is stranded behind the failed node.
• Redundant everything. Nodes, power supplies, fans, back-end paths and JBOF shelf nodes are redundant, and a system has at least two controller nodes.
• Hosts use multipathing across ports on several nodes, so a node or port failure is handled by the host's path failover.

These properties are what let HPE offer the 100% data availability guarantee for the B10000.

Write protection

Writes are protected before they are acknowledged. HPE explains that the B10000 was the first Alletra Storage MP system to use an SSD-based write buffer, and that this delivers the same reliability and low latency as earlier approaches such as NVDIMM, even for latency-sensitive structured data.

Drive protection: RAID 6 and distributed sparing

• RAID 6 only. Every stripe tolerates two simultaneous drive failures. HPE's effective-capacity figures assume a RAID 6 (10+2) layout: 10 data plus 2 parity segments, so parity uses about 17% of the stripe.
• Distributed sparing. Spare space is spread across all drives rather than held on idle spare drives, so rebuilds write to many drives at once and finish faster.
• Wide striping. Each volume spreads across all drives, so a rebuild reads from many drives and the performance impact is spread out.

Data reduction

HPE quotes a 4:1 compaction ratio for effective-capacity planning and backs it with the StoreMore Guarantee. Encrypted, already compressed or media data reduces less, so size from measured data (Chapter 2).

Virtual Copy snapshots

Virtual Copy creates point-in-time snapshots that can be:

• Read/write, for test, development or fast clones.
• Read-only, for consistent backup sources.
• Immutable, which cannot be changed or deleted before their retention expires, even by an administrator. Immutable snapshots are the first line of ransomware recovery.

Snapshots are space-efficient because they share unchanged data with the base volume. Group related volumes into a volume set so they are captured at the same instant (Chapter 7 explains consistency groups).

Remote Copy replication

Remote Copy supports two-data-center (2DC) and three-data-center (3DC) designs, for example synchronous replication to a nearby site plus asynchronous replication to a distant one. Remote Copy over IP (RCIP) uses the optional 10/25GbE replication adapter.

Peer Persistence and Active Peer Persistence

Peer Persistence builds an active-active storage metro cluster from two arrays that replicate synchronously. HPE describes Active Peer Persistence (APP) for the B10000 like this:

• Both the source and target volumes are exported to the hosts, and both arrays service host I/O at the same time. A host can use the local array instead of sending I/O to a remote primary, which lowers latency.
• Manual transparent failover (MTF), or switchover, reverses the replication direction non-disruptively, for example before site maintenance.
• Automatic transparent failover (ATF) redirects host I/O from a failed array to the surviving one, transparently to the hosts. ATF requires an HPE Quorum Witness, which gives independent status for the arrays and prevents split-brain decisions.

Place the Quorum Witness at a third site so that losing either data center, or the link between them, does not also remove the tie-breaker.

Security and ransomware defenses

Non-disruptive operations

• Software upgrades are delivered through DSCC and applied non-disruptively, one node at a time.
• Hardware expansion, adding drives, JBOFs or (on switched systems) controller nodes, happens online.
• Dynamic Optimization changes volume layouts without downtime, and Priority Optimization sets QoS limits so one workload cannot starve another.
• Peer Motion moves volumes in from 3PAR, Primera and Alletra 9000 arrays without application downtime.

Scenario

A bank runs two data centers 20 km apart. The architect deploys a B10000 at each site with Active Peer Persistence over synchronous Remote Copy and an HPE Quorum Witness in a third location. A 3DC design adds asynchronous Remote Copy to a recovery site 600 km away. Immutable snapshots run every hour, and B10000 protection policies send application-consistent backups to StoreOnce. If site A fails, ATF moves host I/O to site B with RPO = 0, while the distant site and StoreOnce cover regional disasters and ransomware.

Common exam traps

• Quorum Witness: needed for automatic transparent failover, not for manual switchover.
• Synchronous replication over long distances: latency makes it impractical; use asynchronous Remote Copy.
• RAID choice: the B10000 does not offer RAID 1, 5 or 10; it uses RAID 6 only.
• Encryption: licensed separately and cannot be turned off once enabled.