Home > Storage > PowerMax and VMAX > Data Protection > Dell EMC PowerMax and VMAX All Flash: TimeFinder SnapVX Local Replication > Performance considerations
Snapshot performance aspects:
I/O profile and snapshot count are both factors of performance.
A snapshot captures a point-in-time image of a TDEV or TDEVs. The image is preserved track-by-track the first time each track is written after the snapshot was taken. Subsequent updates to a track will not require the preservation process for that snapshot.
SnapVX in PowerMaxOS uses redirect-on-write to preserve data for a snapshot. When an update is sent to data that needs to be preserved, the original data remains in place and becomes snapshot data, while the new data is written to another location in the array. If the update is a full 128 KB track, then the update is destaged without any other actions. If the update is less than a full track, then the remainder of the track is read into cache and then the updated version is destaged.
Linked target performance aspects:
It is not necessary to wait for define (nocopy targets) or copy (full copy targets) to complete before accessing the targets. However, there may be some overhead incurred for the first I/O to an undefined target track. This is like accessing a full copy target or clone while copy is in progress. The effect is typically negligible unless there will be heavy I/O to the devices immediately after linking.
Therefore, in some use cases it may be best to wait for the define process to progress before accessing the targets. Even in those use cases, it may not be necessary to wait for the define process to complete. Simply allowing the define process to progress before accessing the targets will reduce or eliminate the chance of accessing undefined target data.
The same rules and considerations apply for snapshot restore operations; Data is available immediately after the restore command completes, there is no need to wait for the copy to complete, and there may be some overhead when accessing tracks that have not completed.
A fully defined nocopy linked target provides the same performance as a fully copied link target. However, some users still choose to use full copy linked targets. The copy rate will dynamically adjust according to system activity to give highest priority to host I/O. Copy times can vary across systems and even within a single system during different periods of activity.
Copy Quality of Service (QoS) allows the user to reduce copy priority below its default value and allow other operations to consume system resources. For more information about QoS, see the Dell EMC VMAX3 and VMAX All Flash Quality of Service Controls for Multitenant Environments White Paper.
Benchmark testing:
Best practice for benchmark testing during SnapVX operations is to use a production-like workload. This will give the most accurate assessment of the performance that can be expected when the application is in production.
The system used in the testing must also be properly sized for the required capacity and snapshot activity, along with any other features that may affect system performance.
Tests using IO levels and change rates beyond that expected in production will not produce realistic results.