[Date Prev][Date Next] [Chronological] [Thread] [Top]

compressors (was: RE 24 testing call (OL 2.4.25))

On Mar 25, 2011, at 1:01 PM, Xin LI wrote:

> Hash: SHA256
> Tests succeeded on FreeBSD/amd64 9-CURRENT.
> By the way, can we consider releasing the tarball with some modern
> compressor, for example xz or bzip2?

First, the best time to choose a compressor is at the first release of the package.  The longer a package been in distribution, the more risk of there is due to breakage of reliance of release conventions.  Switching, I think, would be a really bad idea (I'm not sure you were suggesting this, I'd guess you were suggesting adding an additional compressor).  But even adding a new compressor (as opposed to switching compressors) can lead to breakage in various download, verify, and/or extract scripts.

If I were to select a compressor anew for some new package, I might select bzip2 over gzip.  However, I would be concerned that it's marginal compression gains come at significant computing resource consumption during compression and decompression.  As I find it easier to obtain donations of bandwith than computing resources, I'm more apt to favor gzip over bzip2 even today for source releases.

I also feel that gzip has better proven portability than bzip2 and that bzip2 has better proven portability over xz.  I also note that xz is not available in the default install in all modern UNIX systems, nor is it widely integrated in tar(1) compatible extractors.

I would, in making such a choice for a new package, or subsequently (as with OpenLDAP packages), avoid having multiple compressors if I could because that effectively doubles backup storage/bandwidth requirements for the tarballs.

Lastly, if one could actually reduce bandwidth enough to allow someone to order a smaller pipe, that would be significant.  But 20% or 34% of OpenLDAP tarball bandwidth is not significant to allow any such reduction.  Hell, I doubt our service provider (the ISC, thanks!) would even notice such a reduction.  It's a small percentage of our overall bandwidth usage, and our total a very small amount of their pipe.  At official mirrors, I'm sure we're also in the noise.  And at the downloader, it's only 5MB a pop.

While one can argue that all those 1MB add up, I argue they don't add up to much.

-- Kurt