Which direction should the ratio use?

Original divided by compressed is common, but the convention should always be stated because alternatives exist.

Can compression make a file larger?

Yes. Overhead and limited redundancy can produce expansion.

Is average file ratio the same as overall archive ratio?

No. The overall ratio should normally be calculated from total original and total compressed sizes.

AI Glossary / MLOps and deployment / Beginner

Data compression ratio

Definition

Data compression ratio compares the size of data before and after compression. A common convention divides the original size by the compressed size, so a ratio of 4:1 means the compressed representation occupies one-quarter of the original size.

What it means

Compression ratios summarize storage reduction, but conventions vary. Some reports instead give compressed size divided by original size, while others report the percentage saved. The direction and units must therefore be stated. A ratio also says nothing by itself about whether compression is lossless, how long encoding takes or whether quality remains suitable. Headers, indexes and metadata can materially affect small files, so the measured compressed size should include the components required for actual storage or transmission.

How it works

Measure the original and compressed representations in consistent units. Under the original/compressed convention, ratio R = original size / compressed size. The fraction remaining is 1/R, and the fraction saved is 1 − 1/R. If the compressed representation is larger than the original, R is less than 1 and the percentage saving is negative.
For a collection, dividing total original size by total compressed size gives an overall ratio. Averaging individual file ratios generally gives a different result and can overweight tiny files. Lossy comparisons also require a quality or distortion condition, since stronger information loss can produce a smaller file without delivering an equivalent result.

Three practical examples

A storage team reports the total space saved when compressing an archive.
A media workflow compares file sizes at a specified acceptable quality level.
A data-pipeline benchmark compares compression ratio alongside encode speed, decode speed and memory use.

Think of it this way

Think of packing goods into a smaller container and comparing the occupied space before and after packing. The ratio describes space efficiency, but not how long packing takes or whether anything was discarded. The analogy is limited because digital compression can either preserve every bit or intentionally approximate the content, depending on the method.

Worked example

An original dataset is 120 MB and its complete compressed representation is 30 MB. The original/compressed ratio is 120/30 = 4, conventionally written 4:1. The remaining fraction is 30/120 = 0.25, and the saving is 1 − 0.25 = 0.75, or 75%.
Now consider two files: 100 MB compressed to 50 MB, and 1 MB compressed to 0.1 MB. Their individual ratios are 2:1 and 10:1, whose simple average is 6:1. The actual combined ratio is 101/50.1 ≈ 2.02:1. Aggregating sizes gives the meaningful overall storage result, not the unweighted average of the two ratios.

Illustrative case study

In a fictional document-storage project, a team advertises a strong average ratio based on many tiny repetitive files. When it calculates total storage, a few large already-compressed files dominate and the overall saving is much smaller. The team revises its report to show total bytes and a distribution by file type. It also includes the metadata needed to recover the archive. This case illustrates transparent measurement rather than implying that one reported ratio describes every workload.

Common misconceptions and mistakes

A 4:1 ratio does not mean a 400% reduction; it means 75% smaller under the stated convention. Ratios cannot be compared meaningfully when one excludes metadata and another includes it. A large lossy ratio does not establish good quality. Another mistake is expecting every file to shrink: incompressible or very small data can grow because of headers and coding overhead. A compressed file's size also does not determine peak memory required to compress or decompress it.

Business use

For business decisions, report original size, final size, convention and whether the process is lossless. Include representative file types and total-volume weighting. Compare throughput, latency, memory and quality alongside storage savings. Use the complete deliverable size when estimating bandwidth or retention cost. If a benchmark excludes indexes or dictionaries shared across files, state that choice and account for those resources in the system-level calculation.

Check your understanding

A 200 MB dataset becomes 50 MB. What is its ratio and percentage saving under the original/compressed convention?

Answer: The ratio is 4:1 and the saving is 75%. The compressed representation retains 25% of the original byte count.

Related terms

Lossless compression
Lossy compression
Benchmarks
Compression factor
percentage saving
bitrate
metadata overhead
throughput
rate–distortion tradeoff
benchmark

Learn next

Learn compression benchmarking next. Calculate both per-file and total ratios for a mixed collection, then add encode time, decode time and any quality requirement to the comparison.

Sources

RFC Editor, DEFLATE format specification and representation overhead context: https://www.rfc-editor.org/rfc/rfc1951.html Zstandard project, compression-ratio and speed benchmarking: https://github.com/facebook/zstd Primary format and implementation-source excerpts consulted; full benchmark-specific and editorial review remain pending.

Previous: Data compression Next: Data compression symmetry