Why is *.tar.gz still much more common than *.tar.xz? [closed]

Question

Whenever I see some source packages or binaries which are compressed with GZip I wonder if there are still reasons to favor gz over xz (excluding time travel to 2000), the savings of the LZMA compression algorithm are substantial and decompressions isn't magnitudes worse than gzip.

Answer 1

"Lowest Common Denominator". The extra space saved is rarely worth the loss of interoperability. Most embedded Linux systems have gzip, but not xz. Many old system as well. Gnu Tar which is the industry standard supports flags -z to process through gzip, and -j to process through bzip2, but some old systems don't support the -J flag for xz, meaning it requires 2-step operation (and a lot of extra diskspace for uncompressed .tar unless you use the syntax of |tar xf - - which many people don't know about.) Also, uncompressing the full filesystem of some 10MB from tar.gz on embedded ARM takes some 2 minutes and isn't really a problem. No clue about xz but bzip2 takes around 10-15 minutes. Definitely not worth the bandwidth saved.

Answer 2

The ultimate answer is accessibility, with a secondary answer of purpose. Reasons why XZ is not necessarily as suitable as Gzip:

• Embedded and legacy systems are far more likely to lack sufficient available memory to decompress LZMA/LZMA2 archives such as XZ. As an example, if XZ can shave 400 KiB (vs. Gzip) off of a package destined for an OpenWrt router, what good is the minor space savings if the router has 16 MiB of RAM? A similar situation appears with very old computer systems. One might scoff at the thought of downloading and compiling the latest version of Bash on an ancient SparcStation LX with 32MB of RAM, but it happens.

• Such systems usually have slow processors, and decompression time increases can be very high. Three seconds extra to decompress on your Core i5 can be severely long on a 200 MHz ARM core or a 50 MHz microSPARC. Gzip compression is extremely fast on such processors when compared to all better compression methods such as XZ or even Bzip2.

• Gzip is pretty much universally supported by every UNIX-like system (and nearly every non-UNIX-like system too) created in the past two decades. XZ availability is far more limited. Compression is useless without the ability to decompress it.

• Higher compression takes a lot of time. If compression time is more important than compression ratio, Gzip beats XZ. Honestly, lzop is much faster than Gzip and still compresses okay, so applications that need the fastest compression possible and don't require Gzip's ubiquity should look at that instead. I routinely shuffle folders quickly across a trusted LAN connection with commands such as "tar -c * | lzop -1 | socat -u - tcp-connect:192.168.0.101:4444" and Gzip could be used similarly over a much slower link (i.e. doing the same thing I just described through an SSH tunnel over the Internet).

Now, on the flip side, there are situations where XZ compression is vastly superior:

• Sending data over slow links. The Linux 3.7 kernel source code is 34 MiB smaller in XZ format than in Gzip format. If you have a super fast connection, choosing XZ could mean saving one minute of download time; on a cheap DSL connection or a 3G cellular connection, it could shave an hour or more off the download time.

• Shrinking backup archives. Compressing the source code for Apache's httpd-2.4.2 with "gzip-9" vs. "xz -9e" yields an XZ archive that is 62.7% the size of the Gzip archive. If the same compressibility exists in a data set you currently store as 100 GiB worth of .tar.gz archives, converting to .tar.xz archives would cut a whopping 37.3 GiB off of the backup set. Copying this entire backup data set to a USB 2.0 hard drive (maxing out around 30 MiB/sec transfers) as Gzipped data would take 55 minutes, but XZ compression would make the backup take 20 minutes less. Assuming you'll be working with these backups on a modern desktop system with plenty of CPU power and the one-time-only compression speed isn't a serious problem, using XZ compression generally makes more sense. Why shuffle around extra data if you don't need to?

• Distributing large amounts of data that might be highly compressible. As previously mentioned, Linux 3.7 source code is 67 MiB for .tar.xz and 101 MiB for .tar.gz; the uncompressed source code is about 542 MiB and is almost entirely text. Source code (and text in general) is typically highly compressible because of the amount of redundancy in the contents, but compressors like Gzip that operate with a much smaller dictionary don't get to take advantage of redundancy that goes beyond their dictionary size.

Ultimately, it all falls back to a four-way tradeoff: compressed size, compression/decompression speed, copying/transmission speed (reading the data from disk/network), and availability of the compressor/decompressor. The selection is highly dependent on the question "what are you planning to do with this data?"

Also check out this related post from which I learned some of the things I repeat here.