Atlas running cost optimisations

[zabeen]

Ideas:

• Shared db could be serverless, as metrics show peaks and troughs of activity.
• Matching dbs could be dedicated, if real world usage shows regular activity.
• Run functions on EP2 plan, if search throughput is acceptable (and no out of memory exceptions)
  • Could we separate out search to its own service plan? I think background operations like donor import could be run on a very low plan, or some kind of equivalent to "serverless" where it is mostly dormant.

[mmelchers]

patient with HLA that leads to > 200 thousand donors in a 0 mismatch donor search:

This search would also be the worst 4/8 or 3/6 CBU search.

[zabeen]

App Service Plan

• Large search (over 600K donors) completed on EP2 tier
• WMDA "Peak usage" test set completed in 1hr 15mins on EP2 tier

Conclusion: EP2 tier of elastic service plan should be sufficient to run wmda searches at current levels of usage.

[zabeen]

@seanmobrien please add your notes from your recent investigation into sql db size here, many thanks :)