Performance

Why first access and warm-cache tests need separate results

Understand the value of on-demand access and local caching by checking first reads, repeat work and pre-cached sessions.

See how streaming and caching support the full session

Suite’s first access and repeated access reveal complementary parts of the experience: requested data arrives for new work, and local cached data supports reuse. Capture both, along with deliberate pre-caching, to see how a project behaves from its first open through the working session. Keeping those conditions separate makes your results useful for planning the next workstation or project.

Suite supports direct streaming from the cloud without pre-caching media. Optional pre-caching, automatic client caching and application-generated caches are distinct; planning the latter two does not mean downloading the project before work begins.

Suite documents pre-caching files and folders. That capability is useful, and its preparation should be visible in a comparison. A fast second run is evidence about that run's conditions; it is not a universal throughput or latency result.

Define the task before defining the stopwatch

Choose a user action with a clear endpoint: open a project until the relevant media is usable, play a specified sequence or produce a checked output. Keep the application settings, assets and completion rule constant across observations.

Record both time to useful work and total task duration when they answer different questions. A timeline that becomes usable quickly may still have a long export because of application processing. Combining both into one unexplained number makes the result harder to act on.

For an agent, state whether the metric covers file preparation, the computation itself or the complete job. Include retries and unsuccessful runs rather than timing only the successful attempt.

Label three conditions honestly

  • First observed access: the task has not previously used the selected sample in the recorded session. Record what you know about prior caching; do not claim a perfectly cold state when you cannot establish it.

  • Repeated access: repeat the same task and identify what ran before it. This measures the benefit of the observed prepared state without guessing which cache layer caused it.

  • Deliberately prepared access: record the pre-cache selection, preparation period and readiness checks before timing the task.

Applications and operating systems may also retain data or preparation work. Document the procedure rather than attributing every improvement to Suite. A simple statement of what is known is more useful than an unsupported cache-hit percentage.

Use the same short task in three recorded sessions

Choose a sequence that represents real work: open a project, locate one specified shot, play a defined section, and save a small edit. Keep those actions and the source revision fixed across sessions.

  • Session A — first observed access: Record what is known about prior use on this workstation. Do not label the system completely cold unless that state was actually established. Measure the first usable frame and the complete task.

  • Session B — immediate repeat: Repeat the same sequence on the same host. This captures the combined effect of whatever the client and application retained; it does not isolate a single cache automatically.

  • Session C — planned preparation: If the team uses optional pre-caching, record when it started, when it completed, and how long the same task then took. Include preparation in the total when it blocks the user.

Record network conditions and background work for every session. The useful conclusion may be that direct streaming suits normal work while preparation helps a predictable session. Do not present Session C as proof that pre-caching is required, or Session B as a prediction of every new user’s first open.

Avoid disrupting production to create a benchmark

Use an agreed test environment or disposable sample. Do not clear an active workstation's cache, remove shared files or restart services solely to manufacture a fresh result. If the test needs a specific reset, have the operator approve a supported procedure and document its effects.

Run multiple observations under comparable conditions and retain the range. Note concurrent work, network connection, client version and local storage. A single attractive result is not enough to characterize a team's busiest day.

Publish the conditions with the result

A useful evaluation note says which task ran, how the inputs were prepared, what completed, how long it took and what was not tested. If first access was slow but repeats were acceptable, keep both observations in the report. That distinction may point toward a planned preparation step, a different workload or more investigation.

This article proposes a test method. It does not present measured Suite performance. Use the observations to decide whether the complete workflow meets the team's needs, including the work required before the timed demonstration.

Put the idea to work

Start with your workflow.

Bring shared files into your familiar applications with Suite Managed, or connect your own supported storage with S3 Native. S3 Native requires at least 20 TB of active storage.