What if time did not matter?
Suppose I offer you two ways to store a gigantic object.
Option A occupies a warehouse, but you can retrieve any piece immediately.
Option B fits inside a notebook, but reconstructing the object takes three centuries.
Which representation is smaller?
Easy: B.
Which representation is better?
Now we have to ask what the system is for.
Quick Read
- Compression can reduce storage or transmission while increasing computation.
- Decompression consumes time, memory bandwidth, energy and implementation complexity.
- Stronger compression ratios are not automatically better if encoding or decoding becomes too slow.
- Some compressed formats make random access harder because surrounding state must be reconstructed first.
- Real-time systems have deadlines; a representation arriving too late can be useless even if it is tiny.
- Mobile devices and data centres care about energy as well as bytes.
- The best compression point is therefore an optimisation under multiple constraints, not a contest for the smallest possible file.
The One-Sentence Answer
Compression often saves space by asking the future decoder to perform reconstruction work, so the useful question is not simply how small the representation becomes but where the costs of time, computation, memory, energy and access have moved.
Space and Time Are Different Currencies
A raw object can be expensive to store but cheap to read. A compressed object can be cheap to store but expensive to reconstruct. A carefully designed format can balance both. There is no universal exchange rate because applications value resources differently.
A scientific archive that will be opened once in twenty years may tolerate expensive compression today if storage savings are valuable. A live video call cannot wait minutes for each frame. A game may need assets decompressed within milliseconds. A spacecraft may value transmission bandwidth differently from onboard computation. A battery-powered sensor may care intensely about energy.
Compression becomes engineering when we stop asking for one number.
The Decoder Pays the Bill Later
When a compressor replaces literal data with references, predictions, entropy-coded symbols, transforms or other compact representations, the decoder must reverse those operations.
That reversal may be fast. Modern codecs are often engineered precisely so that decoding is efficient. But it is never metaphysically free. Instructions execute. Memory is accessed. Data moves through caches. Hardware consumes energy. Buffers occupy space.
A smaller disk footprint can therefore produce a larger computational footprint.
Compression Levels Reveal the Bargain
Many practical compressors expose levels or modes. At one end, they prioritise speed. At another, they spend more search, modelling or optimisation effort to obtain a better compression ratio.
This makes the trade-off visible to ordinary users. “Maximum compression” can mean “please spend more computation looking for a smaller representation.” It does not mean “invoke a deeper law of physics.”
Interestingly, compression and decompression costs need not rise together. Some formats deliberately spend substantial work during encoding so that many future decoders can remain fast. This is sensible when an object is compressed once and distributed millions of times.
encode once + decode many times → asymmetric optimisation can make sense
Time Has a Deadline
Now imagine a collision-warning system produces the perfect answer five seconds after the collision.
Its accuracy is irrelevant to the missed intervention.
Information has temporal value. In real-time systems, useful information must often cross a deadline. Compression that saves bandwidth but adds excessive latency can make the total system worse.
This is why “time doesn’t matter” is such a productive thought experiment. It shows us what time was doing in the original problem. Remove deadlines and enormous searches or reconstructions become conceivable. Restore human life, networks, markets, control systems and classrooms, and time reappears as a hard constraint.
Random Access: I Only Wanted Page 73
Suppose an enormous uncompressed record lets you jump directly to a known byte offset. Now suppose a compressed representation depends heavily on earlier symbols or blocks. To recover one small region, the decoder may need to begin from an earlier checkpoint and reconstruct intermediate state.
We saved space, but access became less local.
Designers respond with indexes, blocks, restart points, independently decodable chunks and other structures. Those additions consume some space, but they buy access speed and resilience.
Once again, the smallest representation is not automatically the most useful representation.
A Library Thought Experiment
Imagine compressing an entire library into a miraculous procedural description. You ask for one sentence from one book.
If the only way to obtain that sentence is to regenerate every book from the beginning, the storage achievement may be extraordinary while the retrieval system is dreadful.
Information systems care not only about existence but accessibility.
Bandwidth Can Turn Compression Into Speed
The trade-off can reverse direction.
Suppose a network connection is slow but the receiving processor is fast. Sending a smaller compressed object and spending a little CPU time decoding it may finish sooner than transmitting a huge uncompressed object.
transfer time saved > decompression time added
Compression has then saved both bandwidth and wall-clock time.
This is why resource trade-offs cannot be discussed in isolation. The bottleneck determines which currency is expensive.
Sometimes Compression Saves Energy
It is tempting to say compression always spends energy because computation is required. But moving and transmitting data also consumes energy. In some systems, reducing memory traffic or radio transmission can save more energy than decompression costs.
In another system, an expensive codec may do the opposite.
The correct question is empirical and architectural: where is energy being spent in this workload on this hardware?
The Space–Time Trade Is Older Than Computers
Humans make this exchange constantly.
You can keep a detailed checklist, consuming external space but reducing future recall effort. Or you can memorise the procedure, saving paper while spending learning time and mental capacity. You can pre-cook meals and use refrigerator space to save future preparation time. You can build an index to a book, adding pages so readers spend less time searching.
An index is particularly revealing: we deliberately make the book larger so that information becomes faster to locate.
Sometimes anti-compression is rational.
Education Has a Space–Time Trade Too
A student can memorise a compact formula:
v² = u² + 2as
That is wonderfully economical if the student understands the variables, conditions, relationships and when the equation applies.
Without that decoder, the formula becomes a tiny object requiring expensive reconstruction during an examination. The student stares at it and asks: which value goes where? Is acceleration constant? Which sign should I use? Is this even the right equation?
Expertise reduces future reconstruction cost by spending learning time earlier.
practice now → faster decoding later
Notes Can Be Too Compressed
Students often make notes shorter and shorter because short notes look efficient.
Then three months later:
“Ch 6 — remember exceptions!!!”
The writer at the time possessed the missing context. The future reader—the same human, but with a changed memory state—does not.
This is compression across time with an unreliable decoder.
Good notes preserve enough structure that future reconstruction remains cheap and accurate.
Primary School: Fast to Write, Slow to Understand
Give pupils two instruction cards. One contains complete steps. The other uses unexplained abbreviations. Ask which card is faster to copy and which is faster to follow.
They will discover that saving writing space can create reading time.
Secondary School: Design for Retrieval
Ask students to make two revision sheets: one maximally short and one designed to let their future selves reconstruct a topic accurately after a month. Compare them later.
The best sheet is rarely the one containing the fewest words. It is the one with the best ratio between compactness and reconstruction reliability.
JC and Beyond: Optimise the System, Not the File
At higher levels, compression becomes a multi-objective optimisation problem. Depending on the application, we may care about:
- compressed size;
- encoding throughput;
- decoding throughput;
- peak memory;
- random-access latency;
- energy consumption;
- error resilience;
- implementation complexity;
- hardware acceleration;
- compatibility;
- quality, if loss is permitted.
No single setting maximises all of these simultaneously.
What If We Really Had Infinite Time?
Now return to the original thought experiment.
If computation time were genuinely free and unbounded, we could entertain representations and searches that are completely impractical today. We could spend extraordinary effort trying to discover compact generating descriptions. We could tolerate reconstruction procedures that no real user would wait for.
But infinite time does not repeal the counting limit from lossless compression. There still are not enough short unique descriptions for every long possible string. Unlimited patience can alter what descriptions we can discover or afford to decode; it cannot give every possible object a shorter unique name.
That distinction is beautiful:
more time can improve search for structure more time cannot create nonexistent short descriptions for everything
The Future Is a Receiver Too
Whenever we compress something for archival storage, we are sending a message to the future.
Will future systems possess the codec? The dictionary? The encryption key? The reference files? The hardware? The specification? The cultural knowledge needed to interpret the content?
A tiny archive that nobody can decode is not a triumph of information preservation.