Measure the time of update operations separately

[szarnyasg]

For the interest of analyzing the results, it would interesting to measure the time spent on update operations. E.g. for Umbra this would entail measuring the time spent on 22 operations: 14 inserts and 8 deletes:

insert_nodes = ["Comment", "Forum", "Person", "Post"]
insert_edges = ["Comment_hasTag_Tag", "Forum_hasMember_Person", "Forum_hasTag_Tag", "Person_hasInterest_Tag", "Person_knows_Person", "Person_likes_Comment", "Person_likes_Post", "Person_studyAt_University", "Person_workAt_Company",  "Post_hasTag_Tag"]

delete_nodes = ["Comment", "Post", "Forum", "Person"]
delete_edges = ["Forum_hasMember_Person", "Person_knows_Person", "Person_likes_Comment", "Person_likes_Post"]

It's rather impractical to save the parameters of the entities to the timings file (it would grow massive with billions of operations on SF10,000), so a good schema for storing the results in timings.csv would be:

tool|sf|day|batch_type|q|parameters|time
Umbra|10|2012-12-26|power|insert|Person_knows_Person|123.45
Umbra|10|2012-12-26|power|delete|Forum|67.89

This would allow to see what the decomposition of the writes is (inserts, deletes, precomputations).