From: Tony S. <sch...@bc...> - 2008-07-08 19:08:04
|
On Jul 8, 2008, at 2:55 PM, Evren Yurtesen wrote: > Tony Schreiner wrote: >> On Jul 3, 2008, at 4:07 PM, Evren Yurtesen wrote: >>> Renke Brausse wrote: >>>> Hello Tony, >>>>> I've written before about backups involving very big files >>>>> that seem to execute slowly. >>>>> >>>>> What can be slowing things down so much? Except for this >>>>> operation, everything else runs about as I would expect. >>>> I have no clue what the reason is but I experienced that backups of >>>> large files are much faster with tar over ssh instead of rsync >>>> over ssh. >>>> Not an explanation but maybe this can solve your problem. >>> I believe the reason for this is how rsync works. It normally >>> tries to transfer only the changed parts of the file. This is to >>> save bandwidth, to do this, it has to scan the whole file on >>> both sides (I guess). This is unnecessary unless you are over >>> slow links. You might want to try the whole-file option with rsync: >>> >>> -W, --whole-file copy files whole (w/o delta- >>> xfer algorithm) >>> >>> Please let us know the results, as a side-note if you still want >>> to shrink the transferred file size you can use the ssh >>> compression with -C option of ssh. >>> >>> Thanks, >>> Evren >> Reporting back on this. Using the -W option did not make much >> difference. The dumps in question continue to run for a long time >> and eventually fail with an ALARM. >> For the time being I am excluding the really large files in question. >> I may try tar instead of rsync at some point as was suggested. >> Tony Schreiner > > Are you sure that -W option was active for sure? Did you check with > ps axwww to be sure? or just took it granted that it was active? > > Just asking because I have made several times similar mistake > myself :) > > By the way, if you are using compression, also enabling --checksum- > seed=32761 will make a dramatic speed difference. (although manual > suggests that this can be visible only on 3rd full backup). > > Thanks, > Evren Yes, unfortunately, I looked with ps and the --whole-file option was set. I also have --checksum--seed=32761 set. I'm willing to try again, in case I was hallucinating. Tony |