Investigate possible windows speed issue with large hdf5 files

[Nick Draper]

Check if the issue you noticed actually affects us in the implementation we are using for the event MD workspace file store and file back end implementation.

[Alex Buts]

In [13918]:

refs #3522
Three types of data access strategies evaluated together on machine with disabled prefetching and data write buffer.

[Alex Buts]

In [13921]:

refs #3522
minor variations in output data format to allow simple import and difference between random and non-random data access

[Nick Draper]

Bulk move of tickets to iteration 31 at the iteration 30 code freeze

[Nick Draper]

Moved to iteration 33 at iteration 32 code freeze

[Alex Buts]

refs #3522 different possible file modes description

Do not think, this solves the problem why the performance degradation occurs, but it shows how to avoid this degradation.

Changeset: b244588ae0f388fddacd09f5f6fc6f4abe9040ff

[Alex Buts]

The documents describes experiment on data access speed using three different method. It does not show why the degradation occurs, but finds out how to avoid it (and the price of not-avoidance can be evaluated)

[Alex Buts]

refs #3522 Access time comparison

Reference table

Changeset: 2a799669ac05976282727602d3d267c36fbf19b3

[Alex Buts]

Note to testers.

The tests were implemented for raw hdf, while current Mantid works with nexus in a way, which does not allow easily separate load operations from everything else.

This makes the test largely irrelevant and left here for historical reasons.

[Peter Peterson]

For the reasons mentioned by Alex and the fact that the code is in the last release...I'm going to say this "works".

[Stuart Campbell]

This ticket has been transferred to github issue ​4369