GZipStream and DeflateStream produce bigger files

问题:

I'm trying to use deflate/gzip streams in C# but it appears that the files after compression are bigger than before.

For example, I compress a docx file of 900ko, but it produce a 1.4Mo one !

And it does it for every file I tried.

May be I am wrong in the way I'm doing it? Here is my code :

  FileStream input = File.OpenRead(Environment.CurrentDirectory + "/file.docx");
  FileStream output = File.OpenWrite(Environment.CurrentDirectory + "/compressedfile.dat");

  GZipStream comp = new GZipStream(output, CompressionMode.Compress);

  while (input.Position != input.Length)
      comp.WriteByte((byte)input.ReadByte());

  input.Close();

  comp.Close(); // automatically call flush at closing
  output.Close();

回答1:

Such a big difference seems strange to me, but you should keep in mind that docx is itself compressed in ZIP, so there is no reason to compress it again, results usually are bigger.

回答2:

Firstly, deflate/gzip streams are remarkably bad at compression when compared to zip, 7z, etc.

Secondly, docx (and all of the MS document formats with an 'x' at the end) are just .zip files anyway. Rename a .docx to .zip to reveal the smoke and mirrors.

So when you run deflate/gzip over a docx, it will actually make the file bigger. (Its like doing a zip with a low level of compression over a zipped file with a high level of compression.)

However if you run deflate/gzip over HTML or a text file or something that is not compressed then it will actually do a pretty good job.

回答3:

Although it is true, as others have indicated, that the example files you specified are already compressed - the biggest issue is to understand that unlike most compression utilities, the DeflateStream and GZipStream classes simply try to tokenize/compress a data stream without the intelligence that all the additional tokens (overhead) are actually increasing the amount of data required. Zip, 7z, etc. are smart enough to know that if data is largely random entropy (virtually uncompressable), that they simply store the data "as-is" (store, not compressed), instead of attempting to compress it further.

回答4:

I had the same issue with compressing databases containing jpg data. I tried dotnetzip - a drop in replacement and got decent compression (Supports Compact Framework too!):

MS : 10MB -> 10.0MB
DNZ: 10MB ->  7.6MB

回答5:

I don't think GzipStream and DeflateStream are intended to compress files. You would probably have better luck with a file compressor like SharpZipLib.