Re: [Nfdump-discuss] Compression possibilities of nfdump (and one weird issue)
netflow collecting and processing tools
Brought to you by:
phaag
From: Peter H. <ph...@us...> - 2015-11-21 13:02:32
|
bzip2 is now part of 1.6.14. However, it's not recommended for realtime collecting flows, as it slows down too much in high volume traffic environments. use -z for lzo - still prefered, while collecting. use -j for bzip2 for archiving data or lightly loaded servers Change compression of file: -J <nun> 0: uncompressed, 1: LZO-1, 2: bzip2 - Peter On 06.09.15 13:35, Peter Haag wrote: > Hi, > the compress algorithm was used as best compromise between speed, result and CPU usage. > I also did test bzip2 and - yes - you get more out of disk space, but it's more CPU intensive. > > Anyway, it would be possible to add another compression along the existing one and let the user choose. > So if this is a demand from other people too, we can happily have a lookk together. > > Cheers > > - Peter > > > On 05/09/15 16:01, Miroslav Kratochvil wrote: >> Hello list, >> >> I was wondering whether it is possible to use any better compression for nfdump files than the supplied LZO1X-1. By simple measurements at my site, the files compressed by the nfdump are shrank to around 38% of the original size, while using bzip2 >> (speed of which is still bearable for many environments) gets to around 20% of the original. >> >> For people who need to archive the netflow data, almost two times smaller archives could be a great space/time saver. CPU usage of the compression is IMHO not a real issue on current machines. >> >> Is there any possibility to replace the LZO compression by something else? I'd happily code a patch for bzip2 support -- is that a viable option, or is there some reason not to use this specific compression? (I've seen github migration is underway, >> I'd just send a merge request if it'd been finished :] ) >> >> Thanks a lot in advance, >> -mk >> >> >> PS.: >> Also, on a related note, when trying to code the compression around, I've seen this weird error unexplainable by traditional means: >> >> $ cat < somefile |nfdump -r - >> [... data are OK] >> >> $ bzip2 somefile >> $ bzip2 -c -d < somefile.bz2 |nfdump -r - >> Date flow start Duration Pr[....] >> Read error in file '(null)': Success >> No matched flows >> >> Any explanation? >> >> -- Be nice to your netflow data. Use NfSen and nfdump :) |