# Size offline sync for the depot's morning reconnection

Average traffic hides the burst when many devices reconnect together. Separate small business commands from large files and budget for retries without multiplying work.

By Cobnex editorial. Published 2026-09-10. Updated 2026-09-11.

## Model the burst using actual work units

Consider an illustrative fleet of 80 devices, each returning with 15 reports and four 2 MB photographs per report. That is 1,200 report commands and about 9.6 GB of image data before protocol overhead or retries.

The report API and file-transfer path face different pressure. Scaling a JSON endpoint will not solve a constrained uplink or an image-processing backlog. Measure the stages separately.

Use observed attachment sizes and offline duration when building the real estimate. Averages conceal the worker who has several days of high-resolution photographs waiting on one device.

## Control concurrency at both ends

Limit simultaneous transfers per device and apply server-side admission control across the fleet. Add bounded retry delays with jitter so a brief outage does not cause every device to reconnect in lockstep.

Prioritise according to business value. Sending a small urgent status update before a large optional photograph may be useful, provided the report clearly remains incomplete until its required evidence arrives.

A resumable upload can reduce repeated bytes where the chosen storage service supports it. It still needs a stable association with the report and cleanup for abandoned uploads.

## Spend storage on work that cannot be recreated

Reference manuals and thumbnails may be downloadable again. An unsent inspection photograph may be the only record of a site visit. Treat their local retention priorities differently.

Browser quota estimates are not a reservation of space. Handle write failures and provide a way to remove replaceable cached content without deleting pending reports. Compression decisions should preserve the detail the business needs to inspect later.

A PWA installation does not remove the need to test storage behaviour on supported devices. Capacity planning includes the client's constraints as well as cloud resources.

## Compare total recovery cost

A cheap transfer path can be expensive if support must reconstruct ambiguous submissions. Include duplicate handling, conflict review and device assistance in the operating estimate.

Track bytes retried, queue age and manual recovery frequency after rollout. These measurements show whether to improve compression, upload resumability, scheduling or the command protocol. Buying more server capacity is useful only when the server is the limiting stage.

## Sources

- [MDN: browser storage quotas and eviction](https://developer.mozilla.org/en-US/docs/Web/API/Storage_API/Storage_quotas_and_eviction_criteria)