Let assume I need to send some data from one computer to another, over a pretty fast network... for example standard 100Mbit connection (~10MB/s). My disk drives are standard HDD, so their speed is somewhere between 30MB/s and 100MB/s. So I guess that compressing the data on the fly could help.

But... I don't want to be limited by CPU. If I choose an algorithm that is intensive on CPU, the transfer will actually go slower than without compression.

This is difficult with compressors like GZIP and BZIP2 because you usually set the compression strength once for the whole transfer, and my data streams are sometimes easy, sometimes hard to compress--this makes the process suboptimal because sometimes I do not use full CPU, and sometimes the bandwidth is underutilized.

Is there a compression program that would adapt to current CPU/bandwidth and hit the sweet spot so that the transfer will be optimal? Ideally for Linux, but I am still curious about all solutions. I'd love to see something compatible with GZIP/BZIP2 decompressors, but this is not necessary.

So I'd like to optimize total transfer time, not simply amount of bytes to send.

Also I don't need real time decompression... real time compression is enough. The destination host can process the data later in its spare time. I know this doesn't change much (compression is usually much more CPU-intensive than decompression), but if there's a solution that could use this fact, all the better.

Each time I am transferring different data, and I really want to make these one-time transfers as quick as possible. So I won't benefit from getting multiple transfers faster due to stronger compression.


  • 1
    +1 interesting and potentially useful idea. If you ever find some good ways to do this, I'd like to hear about it. – David Cary Mar 16 '11 at 16:28

This is a current subject of research - primarily in the area of sensor networks where the goal is to minimise power usage, rather than maximise throughput. The principle of adaptive compression is the same however.

Here is a recent paper from a professor at USC.

Maybe you could have a go at implementing his algorithm? I'm sure there would be many people interested in a good implementation.

  • I'd love to, if I could... the usual human time constraints apply here though. Still I can be first to test it ;-) Thank you for links to papers, I might try to find other people that might be interested. – liori Jun 17 '10 at 18:25

Hum, this question is older than one year now, so this may pass unnoticed:

Anyways, Google has recently published Snappy that may be exactly what you're looking for, namely:

[...] It does not aim for maximum compression, or compatibility with any other compression library; instead, it aims for very high speeds and reasonable compression. For instance, compared to the fastest mode of zlib, Snappy is an order of magnitude faster for most inputs, but the resulting compressed files are anywhere from 20% to 100% bigger [...]

It's implemented in C++ with bindings available for C and a range of other languages.

  • Oh, no, it won't be unnoticed, at least by me ;-) And it seems it is not adaptive in the sense I look for--it is just fast. – liori Aug 22 '11 at 13:11

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.