John Jolet wrote: > > On Jan 24, 2006, at 9:10 PM, Ow Mun Heng wrote: > >> On Tue, 2006-01-24 at 17:23 +0000, Francesco Riosa wrote: >> >>> Jeff wrote: >>> >>>> Hey guys. >>>> >>>> I've got this big fat backup server with no space left on the hard >>>> drive >>>> to store a tar file. I'd like to pipe a tar through ssh, but not sure >>>> what the command would be. Something to the effect of: >>>> >>>> # cat /var/backup | ssh backup.homelan.com 'tar data.info.gz' >>>> >>>> So that, the data is actually being sent over ssh, and then >>>> archived on >>>> the destination machine. >>>> >>> tar -zcf - /var/backup | ssh backup.homelan.com "( cat > >>> data.info.gz )" >>> >> >> There's another way. This assumes your originating server's CPU is >> slow/precious and you have a 16 way node on a backup server (HAHA!!) >> >> tar cf - /var/backup | ssh backup.homelan.com "gzip -c > >> filename.tar.gz" >> >> But you transfer the stream uncompressed, so more bits get transferred. >> > you're kidding, right? Unless you've got a PII on the originating > end and are using gigabit ethernet between the two nodes, compressing > the data before transmission will almost always be faster. I tested > this scenerio extensively about 3 years ago, using linux, aix, and > solaris hosts. In no case was transferring uncompressed data faster > than compressing (at least to some degree) the data on the > originating server. And frankly, no matter what you do...wouldn't > you hope ALL the bits get transferred? :)
I read something some time ago that suggested if you transfer a compressed file over a compressed SFTP connection, for example, that it would take longer to transfer the data versus if only the data or the connection was compressed. The reason, as I recall, had to do with compressing already compressed data--this apparently created some overhead on the connection. Did you look at this situation in your tests? If so, what were the results? -- gentoo-user@gentoo.org mailing list
