feasibility of buffering multiple volumes

[toloudis]

Use Case

it would be great to be able to do real time playback of time series data but buffering multiple volumes is memory intensive.

Can we detect anything about a user's system resources (cpu and/or gpu) and provide some automatic buffering capability?
Can we allow power users some control over pre-buffering many times in a time series?

Solution

Prototype ui widgets to let user say how many times to prebuffer. Include some indicator of buffering progress and range of times in the buffer (say times 30-60 out of a 0-500 range). Determine what things we can measure about a user's system resources from in-browser.

[toloudis]

Requirements for a good caching system for these images:

• bounded size -- but could be adaptive based on auto detection of memory -- or a build-time configuration setting -- or a user setting
• fetch and cache times around the current time. users are most likely to advance in time in this way.
• if user jumps far away from cached values, we need to potentially discard existing, cancel anything in-flight, and begin requesting around the new current time
• we are assuming whole multichannel volumes but this could also apply to single slices or other data types.

[toloudis]

There is a deprecated but probably-still-working performance.memory api available in chrome and edge but not firefox nor safari

[frasercl]

We can play through time now! It works ok!