Estimate the transfer with your files and your observed throughput.

Count what really needs to travel

Make two inventories: the items to send to work and those to retrieve at the end. Sources, the resources needed by the software and the retained results do not necessarily have the same volume. Also separate the files you can prepare before the rental from those that will only exist after the compute.

Note the number of files and their total size in bytes when your tool provides it. Distinguish the size of the data from the space used on disk. If you are preparing an archive, use its actual size for the transfer of that archive. Packing and extracting it take time and may require extra space: check these separately, without assuming any compression gain.

Set aside unnecessary copies after verifying their role. A public, retrievable cache does not always have the same priority as a unique original or project settings. Sorting the items to keep is detailed in the backup guide.

Put volume and throughput in the same units

In this guide, one byte equals eight bits. Decimal prefixes give 1 MB = 1,000,000 bytes and 1 GB = 1,000,000,000 bytes. Binary prefixes give 1 MiB = 1,048,576 bytes and 1 GiB = 1,073,741,824 bytes. The corresponding English notations are MB, GB, MiB and GiB; NIST documents these distinctions.

Mbit/s and MB/s therefore do not refer to the same quantity. A throughput of 80 Mbit/s corresponds arithmetically to 10 MB/s. This conversion does not promise that the application will actually copy ten megabytes of files per second. Always note the unit displayed, especially the difference between bits and bytes.

Do a small round trip before the big folder

Once your environment and transfer method are available, send a small permitted file, retrieve it elsewhere and open that copy. This first check validates the path, the access rights and the destination. It does not yet properly measure the time for a large folder.

Then choose a representative group, large enough to observe the transfer beyond its startup. Use the same tool, the same destination and a composition close to the real folder. If you have many small files, test a group of those files, not just one big video. Note the volume actually copied and the time elapsed until the end of the transfer, including enumeration and pauses.

Calculate the average as volume divided by time. Repeat at another time if the result varies greatly and keep both observations. Measure upload and retrieval separately: an observation in one direction is not a measurement in the other. A network test to another destination is not your file transfer either. ESnet distinguishes network throughput testing from the limit it places on the actual transfer.

Calculate a duration, then check what it covers

With consistent units, duration in seconds = volume in bytes ÷ throughput in bytes per second. For a decimal volume in GB and a throughput in MB/s, this becomes duration in seconds = GB × 1,000 ÷ MB/s. Then divide by 60 to get minutes. The throughput must be strictly positive; if there is no usable measurement, leave the duration as unknown.

For example, a group of 800 MB copied in 100 seconds gives an average of 8 MB/s. This figure describes only that observation. If your timing already includes pauses and startup, do not add an arbitrary extra allowance a second time for those same steps. Instead, set aside separate rows for actions that were not measured: preparing the archive, opening the copy or redoing part of the transfer.

A cautious scenario may use the lowest of several representative averages. Note which ones and their conditions. Choosing a lower value is a planning assumption, not a guarantee that the throughput will stay above it.

Worked example: sending 6 GB and retrieving 1.2 GB

Suppose a project with files totaling 6 GB to send, then 1.2 GB of results to retrieve. For this entirely hypothetical example, the upload test gives 8 MB/s and the retrieval test 3 MB/s. All volumes are decimal. These values do not come from a test on a BriefGPU rental.

The upload would take 6,000 ÷ 8 = 750 seconds, or 12 minutes 30 seconds. The retrieval would take 1,200 ÷ 3 = 400 seconds, or 6 minutes 40 seconds. The two transfers therefore amount to 19 minutes 10 seconds if the average speeds used hold up. The table adds control and buffer assumptions decided for the project.

Test the effect of another assumption

If the retrieval speed in the example drops to 1.5 MB/s, the 1,200 MB take 800 seconds, or 13 minutes 20 seconds. The time for this step doubles. The total, with the other assumptions unchanged, rises to 1 hour 15 minutes 50 seconds. This shows you which part of the schedule depends on the network and which part depends on your own controls.

If the announced volume were 6 GiB and not 6 GB, the upload at 8 MB/s would take about 805 seconds, or 13 minutes 25 seconds. The difference here comes from the units, with no change in the network. For an accurate calculation, start from the number of bytes rather than a rounded value displayed by a file explorer.

Finally, if the actual folder does not have the same composition as the sample, take a new measurement. A simple multiplication does not make up for a test that does not represent your copying method.

Put the transfer in the right place on the schedule

The upload done after the server is made available counts as part of preparing the rental. The retrieval and final verification belong to the export. Any preparation finished earlier on your own computer stays separate. Do not count the same interval twice if a transfer and another action happen in parallel.

The total in the example includes neither installation, nor computation, nor waiting for client feedback. Add these items to your full schedule before comparing the 3-, 7-, and 30-day plans. The method makes no claim about the server's delivery time, storage capacity, or network characteristics.

Avoid putting off any retrieval until the last moment. An intermediate copy and a first file opened from the destination reveal obstacles early. The transfer's completion message does not replace checking the expected files.