How to improve speed of a single file transfer over an unstable link?

[Scott C. Kennedy]

Thank you for the idea, I just checked and....

The cygwin version of rsync on my parent's machine is 2.6.6 and
'--append' is a 2.6.7 feature! :(

So, until cygwin updates, I came up with this idea....

1)'split' the file into 100 MB chunks on both ends.
2) rsync the two directories, thus the file date & size for MOST files
will be correct and then do an '--inplace' to continue updating.

Thoughts?

Scott
PS> the split just died since the link dropped again. :( The issue is
the cabling in my father's house is poor, thus the cable modem keeps
error-ing out and they are 3,210 miles away from me right now.

Paul Slootman wrote:
> On Fri 29 Dec 2006, Scott C. Kennedy wrote:
>   
>> Thus, I've scripted the following script 'get_me.sh'
>>
>>    #!/bin/sh
>>    mv .file.bkf.* ./file.bkf
>>    rsync --timeout 90 user at remote:/dir/file.bkf ./file.bkf
>>    ./getme.sh
>>
>> So, the script moves the temp file created by rsync onto the file itself,
>> then calls rsync to continue sync'ing, and then after rsync losses it's
>> connection, the script calls itself and the cycle starts again.
>>     
>
> You do know about the --partial option? That basically takes care of
> this...  although I'm wondering why your rsync doesn't delete the
> tmpfile after the transfer is interrupted.
> Using --inplace may also be useful.
>
>   
>> Not very elegant but it's working. Sort of. I'm now starting to decrease
>> the overall throughput of the transfer since I keep checking to make sure
>> that the data is still the same on both side, so here's the question...
>>     
>
> rsync will check the existing data upon the start of each transfer,
> unless...
>
>   
>> Would the "append" flag work well for this situation? I'd normally try a
>> few tests myself, but according to my data, it'll be at least another 4
>> days until the file is finished, and my parent's leave in 5 days. So, I'm
>> a little hesitant to "experiment" on the transfer in progress.
>>     
>
> the --append option will assume that the partial data there is correct,
> only too short. This is a good idea, as it saves reading the already
> transported data, saving time. And --append implies --inplace.
>
>
> Paul Slootman
>   

-------------- next part --------------
HTML attachment scrubbed and removed