Compression is a confusing issue. Sequence files that are in block format are always split table regardless of what compression for the block is chosen.The Programming Hive book has an entire section dedicated to the permutations of compression options.
Edward On Mon, Nov 5, 2012 at 10:57 AM, Krishna Rao <krishnanj...@gmail.com> wrote: > Hi all, > > I'm looking into finding a suitable format to store data in HDFS, so that > it's available for processing by Hive. Ideally I would like to satisfy the > following: > > 1. store the data in a format that is readable by multiple Hadoop projects > (eg. Pig, Mahout, etc.), not just Hive > 2. work with a Hive external table > 3. store data in a compressed format that is splittable > > (1) is a requirement because Hive isn't appropriate for all the problems > that we want to throw at Hadoop. > > (2) is really more of a consequence of (1). Ideally we want the data stored > in some open format that is compressed in HDFS. > This way we can just point Hive, Pig, Mahout, etc at it depending on the > problem. > > (3) is obviously so it plays well with Hadoop. > > Gzip is no good because it is not splittable. Snappy looked promising, but > it is splittable only if used with a non-external Hive table. > LZO also looked promising, but I wonder about whether it is future proof > given the licencing issues surrounding it. > > So far, the only solution I could find that satisfies all the above seems to > be bzip2 compression, but concerns about its performance make me wary about > choosing it. > > Is bzip2 the only option I have? Or have I missed some other compression > option? > > Cheers, > > Krishna
