Start from 0.7.x version, dictionary is enabled by default for all no-numeric dimensions, and there's no control option for user.
Please upgrade to 0.7.x if you are still using 0.6 version which we no longer supported. Thanks. Luke Best Regards! --------------------- Luke Han On Fri, Jul 10, 2015 at 12:03 AM, Sarnath <[email protected]> wrote: > Hi Abhilash, > So Dictionary is not built by default? We have to explicitly ask Kylin to > do that? > > On Thu, Jul 9, 2015 at 9:27 PM, Abhilash L L <[email protected]> > wrote: > > > @Sarnath > > There is an option to use dictionary for a particular dimension. > > If selected, during building the base cuboid, it use the dictionary to > map > > the value to an id. > > > > Please note that the row key is composed of multiple such dimensions. > > > > Someone please correct me if wrong, I think there is a limit of 2 million > > on cardinality on the dictionary. (as per 0.7.1) > > > > Regards, > > Abhilash > > > > On Thu, Jul 9, 2015 at 9:09 PM, Sarnath <[email protected]> wrote: > > > > > Hi Luke, > > > That helped. Thanks... So, the RowKey in HBASE uses the dictionary-ID > > for a > > > dimension's value instead of using the actual value.. Is that correct? > > > > > > On Thu, Jul 9, 2015 at 7:23 PM, Luke Han <[email protected]> wrote: > > > > > > > Hi Sarnath, > > > > You are welcome. > > > > Kylin has dictionary for all dimensions, so id vs name is no > > matter, > > > > just pick up one or both best for your case. > > > > > > > > So memory consumption between denormalized and normalized should > > just > > > > be a litter bit different. > > > > > > > > Hope such explain could help you. > > > > > > > > Thanks. > > > > > > > > > > > > Best Regards! > > > > --------------------- > > > > > > > > Luke Han > > > > > > > > On Thu, Jul 9, 2015 at 9:20 PM, Sarnath <[email protected]> wrote: > > > > > > > > > Hi Luke, > > > > > Thanks. This is good to know. > > > > > Also, Will use of denormalized tables result in more memory > > consumption > > > > in > > > > > kylin than maintaining a normalized one? > > > > > E.g. It is cheaper to store a country ID than country name. > > > > > (Or) will kylin build an internal set of IDs and not store the > > strings > > > > > directly? > > > > > Thanks > > > > > Best, > > > > > Sarnath > > > > > On Jul 9, 2015 5:40 PM, "Luke Han" <[email protected]> wrote: > > > > > > > > > > > HI Sarnath, > > > > > > Kylin supports denormalized table, you can generate one flat > > file > > > > and > > > > > > just use it as source to build cube. > > > > > > > > > > > > To leverage Star Schema is easy for people to understand data > > > model, > > > > > > most of cases are using star schema to store data and so on. > > > > > > > > > > > > Thanks. > > > > > > > > > > > > Luke > > > > > > > > > > > > > > > > > > Best Regards! > > > > > > --------------------- > > > > > > > > > > > > Luke Han > > > > > > > > > > > > On Thu, Jul 9, 2015 at 8:01 PM, Sarnath <[email protected]> > > wrote: > > > > > > > > > > > > > Hi, > > > > > > > Can kylin work with denormalized tables? Or is star schema > > > mandatory? > > > > > > > Why I ask this is : in hadoop world most tables are > > denormalized.. > > > > > > > Is there any motivation behind star schema? > > > > > > > Best, > > > > > > > Sarnath > > > > > > > > > > > > > > > > > > > > > > > > > > > >
