Java - Google App Engine - modelling graph structures in Google Datastore -


google apps engine offers google datastore nosql database (i think based on bigtable).

in application have social-like data structure , want model in graph database. application must save heterogeneous objects (users,files,...) , relationships among them (such user1 owns file2, user2 follows user3, , on).

i'm looking way model typical situation, , thought 2 families of solutions:

  1. list-based solutions: object contains list of other related objects , object presence in list relationship (as google said in jdo part https://developers.google.com/appengine/docs/java/datastore/jdo/relationships).

  2. graph-based solution: both nodes , relationships objects. objects exist independently relationships while each relationship contain reference 2 (or more) connected objects.

what strong , weak points of these 2 approaches?

about approach 1: simpler approach 1 can think of, , presented in official documentation but:

  • each directed relationship make object record grow: there limitations on number of possible relationships given instance object dimension limit?
  • is jdo feature or datastore structure allows approach naturally implemented?
  • the relationship search time increase list, solution suitable large (million) of relationships?

about approach 2: each relationship can have higher level of characterization (it object , can have properties). , think memory size not google problem, but:

  • each relationship requires own record, search time each related couple increase total number of relationships increase. suitable large amount of relationships(millions, billions)? i.e. google have tricks search among records if structured? or in situation in if want search friend of user1 called user4 have wait seconds?
  • on other side each object doesn't increase in dimension new relationships added.

could me find other important points on 2 approaches in such way chose best model?

first, search time in datastore not depend on number of entities store, on number of entities retrieve. therefore, if need find 1 relationship object out of billion, take same time if had 1 object.

second, list approach has serious limitation called "exploding indexes". have index property contains list make searchable. if ever use query references more property, run issue - google understand implications.

third, list approach more expensive. every time add new relationship, rewrite entire entity @ considerable writing cost. reading costs higher if cannot use keys-only queries. object approach can use keys-only queries find relationships, , such queries free.

update:

if relationships directed, may consider making relationship entities children of user entities, , using object id id relationship entity well. relationship entity have no properties @ all, cost-efficient solution. able retrieve objects owned user using keys-only ancestor queries.


Comments

Post a Comment

Popular posts from this blog

javascript - jquery or ashx not working -

i using ajax file upload control. it creating html: <form action="/pages/handlers/asyncfileuploadhandler.ashx" method="post" enctype="multipart/form-data" target="ajaxuploader-iframe-1397144545"> <input type="hidden" name="type" value="mindvoice"> <input type="file" id="imagefile" style="display: block; width: 80%; margin-right: auto; margin-left: auto;"> </form> and target is: <iframe width="0" height="0" style="display:none;" name="ajaxuploader-iframe-1397144545" id="ajaxuploader-iframe-1397144545"></iframe> the form submitted using .submit() . my ashx handler: <%@ webhandler language="c#" class="asyncfileuploadhandler" %> using system; using system.web; using nettpals.core; public class asyncfileuploadhandler : ihttphandler { public void processre...
Read more

opencv - DataType<cv::detail::deriv_type>::depth what is it used for -

i wondering why line used in lucas kanade tracker in opencv: datatype<cv::detail::deriv_type>::depth can explain me? in opencv, depth of mat refers type of data contained in mat 's data buffer. represented integer values correspond given data type. these integers commonly abstracted appropriate macro definition (e.g. uchar data represented macro cv_8u ). cv::datatype type-traits class provides method obtain corresponding integer value without having memorize macro means data type. there few cases user code needs use datatype::depth . more common datatype::type . a simple example shows 1 possible use of datatype::depth : cv::mat uchar_data = cv::mat::ones(3, 3, cv_8uc1); cv::mat float_data; uchar_data.convertto(float_data, cv::datatype<float>::depth); // ^^ equivalently replaced // cv_32f macro float_data.at<float>(0,1) += 0.5f; std::cout << float_data << std::endl;...
Read more

python 3.x - Mapping specific letters onto a list of words -

can please suggest approach write piece of codes automatically maps letter in letter_str onto dic_key (a dict key string type contains dashes match length of words in word_lst)? so, mapping occurs if letter appears in every words in list @ same position no matter how many words in list. if no letter appears @ position words in word list new_dic_key '----'. please see examples below thanks word_lst = ['ague', 'bute', 'byre', 'came', 'case', 'doze'] dic_key = '----' letters_str ='abcdefghijklmnopqrstuvwxyz' new_dic_key = '---e' if word_list = ['bute', 'byre'] new_dic_key = 'b--e' or word_list = ['drek', 'drew', 'dyes'] new_dic_key = 'd-e-' if words in word_list of same length code give want: word_list = ['drek', 'drew', 'dyes'] cols = [] in range(len(word_list[0])): cols.append([]) word in ...
Read more