Much appreciated, Zooko. I just reported to Allmydata.com support a significant slowdowns I am experiencing when backing up over longer periods of time (10-15 hours). I wanted to make sure that was not a network issue before reporting it on this list.
> In fact, that graph reminds me of an even *easier* hack, just set the > segment size back to a few MiB instead of its currently value of 128 > KiB, and see if that significantly improves performance from Andrej's > perspective! Please let me know when there is something to test. Also, I'm not clear on what 'tahoe backup' does when only a small part of the file changes - does it upload only changed blocks, or the whole file? I'm sure this is documented somewhere, but unfortunately I am not fammiliar with the terminology used in many places. What happens to the files that are moved/renamed? Cheers, Andrej On Tue, Mar 10, 2009 at 4:04 AM, zooko <[email protected]> wrote: > On Mar 4, 2009, at 20:23 PM, Andrej Falout wrote: > > > In your plans, is there something you can do to improve upload > > speed? Directory creation speed? > > Thanks for the feedback! Again, I'm really glad to have an end-user > on this list reminding the developers of what he wants. > > Let's prioritize faster uploads and downloads for tahoe-1.4. This is > the ticket describing what we already know about it: > > http://allmydata.org/trac/tahoe/ticket/392 # pipeline upload segments > to make upload faster > > It would actually be a reasonable task for a newcomer to tackle. To > real solve it in the best possible way is equivalent to > reimplementing TCP on top of foolscap messages, or else switching > from foolscap to HTTP for storage protocol, but to do some reasonable > hack that makes it, let's say "twice as fast", for typical > situations, should be quite doable. > > See these graphs of how throughput has *dropped* dramatically as > we've changed Tahoe to make it better along other dimensions: > > http://allmydata.org/tahoe-figleaf-graph/hanford.allmydata.com- > tahoe_speedstats_rate.html > > In fact, that graph reminds me of an even *easier* hack, just set the > segment size back to a few MiB instead of its currently value of 128 > KiB, and see if that significantly improves performance from Andrej's > perspective! > > Regards, > > Zooko > - Show quoted text - > _______________________________________________ > tahoe-dev mailing list > [email protected] > http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev > -- Andrej Falout AU: +61 (410) 463 735 NZ: +64 (21) 0256 6825 US: +1 (360) 488 0970
_______________________________________________ tahoe-dev mailing list [email protected] http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev
