PowerMax 2500 and 8500: Unmanaged Snapshots

Unmanaged Snapshots in PowerMax 2500 and 8500
Local Replication (LREP)

SnapVX has been rearchitected in V4. V4 defaults to copy on a read and copy on a write when intercepted. V3 and before did not default to copy on read.

Unmanaged Snapshots: 

• Like a VMAX3, a PowerMax 2500 or 8500 maintains Target data even after it is Unlinked from the Source devices.
• Unmanaged snapshots are images kept internally by the PowerMax to maintain dependencies between a Source and Target after they are Unlinked.
• There are fundamental differences between the way LREP/VP is implemented in PowerMax 2500 or 8500 compare with the older VMAX All Flash/VMAX3/Powermax Arrays.
• In PowerMax 2500 or 8500, target devices do not go through the define process
• The VMAX3 define process would consume full Track ID (TID) metadata for a target device, making overprovisioning expensive
• With PowerMax 2500 or 8500, it was decided to do away with track-level sharing and use a device-level sharing scheme now (while also keeping target devices undefined)
  • This also improved performance on undefined targets
• So, in PowerMax 2500 or 8500, targets only get allocations when they are written to (not as part of the define process)
  • They derive their data from the point-in-time image, which they were previously linked to.
  • This relationship remains after they are unlinked
  • Even in VMAX3, data integrity was maintained on a target device after it was unlinked (it was done at the track level)
• If getting an IO to an unmanaged target device, like every IO, code checks if it must intercept
  • This is done early in the IO flow
  • Read IOs to target tracks that have not yet been copied, generate SNAPVX_COPY_ON_READ_INTERCEPT if it was asked to do so.
    • For MF IOs, there are a few cases that generate COPY ON READ INTERCEPTS
      • Asked to do COPY_INDIRECT_ON_READ
      • It is a special MF 0x96, op 0x46 command that forces a copy on read
      • It is an SRDF R2 SRDF/A device, and all reads are set COPY ON READ (Example: RDC scans do this because a request is sent to the io machine to read the data from the remote site to compare. This causes a read intercept, forcing the copy of data)

General example with Unmanaged Targets:

• The sequence A -> s1 -> B represents a Target device (B) linked to a snapshot (s1) of the source device (A)
• This relationship remains even after unlinking the Target (B) and terminating the snapshot (s1)
• As writes occur to the source device (A), RDP is pushed to the snapshot (s1) such that it continues to accumulate allocations
• These RDP can be shared across multiple snapshots and used by multiple Targets (even snapshots that are still Active with Linked Targets)
• Data written to the Target (B) are the only allocations that show up on the Target (B)
• Running "free-all" on the Target (B) removes the internal relationship and dependency on the terminated snapshot (s1)
• Similarly, linking the target to a different snapshot removes the internal relationship and dependency on the terminated snapshot (s1)