The replication in 3PAR Arrays has always been mediocre. In the older versions of 3PAR Inform OS if you choose ‘sync replication’ for a single remote copy group, you could not use ‘a sync’ for any other remote copy groups.
This feature was addressed in a newer version of 3PAR Inform OS, however your lowest RPO using ‘a sync’ was bottlenecked at 15 minutes regardless of available bandwidth.
With the release of the HP 3PAR 20000 Series, comes a new feature which is streaming ‘a sync’ replication.
What Is Streaming ‘A Sync’ Replication
Essentially, if you have the bandwidth and cache available the source 3PAR will stream replication across to the target 3PAR reducing your RPO below 15 minutes. I like to think of it as a best endeavours.
When designing a replication infrastructure it’s important to know the transport method as well as the thresholds in terms of bandwidth and latency between source and target arrays. This ensures that you are not only within a supported SLA, but also to ensure that write performance of the source array is not effected.
The table below shows supported thresholds.
The source array uses a local cache to maintain host write transactions in memory. A concept known as ‘delta sets’ are used.
Source 3PAR Array
- I/Os are transferred from primary array to secondary array as part of a delta sets
- I/Os on the primary array that belong to a particular remote copy group are grouped together into delta sets
- A delta set is made up of sub-set of I/Os, where sub-set represents I/Os owned by a remote copy group on each given node
Target 3PAR Array
- A delta set is applied on the secondary RC volume group only after:
- The entire delta set has been received in the secondary array cache
- And the previous sets that this delta set depends upon have completed.
- A secondary RC volume group is always in a crash consistent state, before or after the application of a delta set. It is not crash consistent during the application of a delta set.
- If the delta set fails to apply on the secondary volume then the group stops and a fail back to the last coordinated snapshot is required
What About Write Bursts?
A write burst is when the array receives a significant number of writes which could last for a few minutes. If the inter site link between source and target array is sufficient this has no impact.
It is when the inter site link cannot cope or the write cache gets filled then the source 3PAR will choose a random remote copy group to stop and a snapshot is taken.
Note: You have no control over which remote copy group
Once stopped these groups will start again at the next sync period.
This is a great feature set being added to the 3PAR 20,000 Series. I’m sure when the next .1 release update is received you will be able to select which remote copy groups you would want to stop either due to a write burst or cache overflow.
With most 3PAR updates, I expect the streaming ‘a sync’ replication to find its way into the 7×00 series within a short period of time.