On Thu, Feb 2, 2012 at 5:35 PM, 胡瀚森 <softrank.net@xxxxxxxxx> wrote: > Thanks a lot. > But I still don't know what does the word "probabilistically" in the sentence. >>>...thus defining a sequence of candidates for each replica rank that are [[probabilistically]] independent of others' failures. > That seems to be expressing that there is a chance to be independent, > not 100% independent, and in some certain conditions they could be > dependent. Am I right? > If I'm right, what kind of situation can it be? Sage can correct me if I'm wrong, but I think the use of "probabilistically" there is more a reminder of the domain you're working in than a word that changes the meaning. CRUSH is a probabilistic algorithm and the choice of replacements is probabilistically independent — but that doesn't mean that the choice of replacements for your storage replicas is necessarily independent if there are physical failure correlations that aren't reflected in your CRUSH model. Does that make sense? > On Fri, Feb 3, 2012 at 3:43 AM, Gregory Farnum > <gregory.farnum@xxxxxxxxxxxxx> wrote: >> On Thu, Feb 2, 2012 at 12:35 AM, 胡瀚森 <softrank.net@xxxxxxxxx> wrote: >>> Hi everyone. >>> >>> I've received a nice and welcome response last time. Thank you. I've >>> got some other questions this time. >>> >>> thesis can be found: http://www.ssrc.ucsc.edu/Papers/weil-sc06.pdf >>> >>> at page 4, chapter 3.2.2 Replica Ranks, under Figure 2. >>>>>...thus defining a sequence of candidates for each replica rank that are probabilistically independent of others' failures. >>> Here, why "probabilistically" instead of "certainly"? and what's >>> others' failure? >> >> The statement means that the sequence of replacement choices for >> position 2 is independent of whether position 1 has failed — the >> current node occupying position 1 isn't part of the function's input; >> there isn't a roundabout way that it influences the input, etc. >> >>>>>...making it unweildly for parity schemes. >>> Here, what does "unweildly" mean? is it "unwieldy" or something else? >> >> Yep, that's a typo. It should say "unwieldy"! :) >> -Greg -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
