Skip to content

Latest commit

 

History

History
40 lines (27 loc) · 2.59 KB

MDB.md

File metadata and controls

40 lines (27 loc) · 2.59 KB

Instructions for replicate the experiments in MillenniumDB

Compile MillenniumDB

Database creation

  • Transform the N-Triples File downloaded from Figshare to the text format for MillenniumDB:

    • python3 scripts/nt_to_mdb.py truthy_direct_properties.nt truthy_direct_properties.mdb

  • Execute the bulk loading:

    • build/Release/bin/create_db truthy_direct_properties.mdb tests/dbs/wikidata

Execute query set

Before executing the script you need to edit some paths to match the correct folders on your machine:

MDB_PATH = f'/home/user/MillenniumDB' # where you cloned MillenniumDB repo
OUTPUT_PATH = f'/home/user/results'   # results and logs will be written here, folder must exist

Then you can run the script choosing the desired parameters:

  • python3 scripts/benchmark_mdb.py [query_file] [ANY|SIMPLE|TRAILS|ALL|ALL_COUNT] [bfs|dfs] [cache|naive] [btree|trie]

    for example:

    • python3 scripts/benchmark_mdb.py queries/set_I.txt ANY bfs cache btree

Considerations for the parameters

  • The param block [ANY|SIMPLE|TRAILS|ALL|ALL_COUNT] selects the desired semantics of returned paths. Here ALL stands for ALL SHORTEST. The semantics of ANY will depend on the selected algorithm (see [bfs|dfs] below).

  • The param block [btree|trie] selects the data access method. The btree option uses the disk data stored in B+trees and buffered into main memory. The trie option stores the data in main memory using a compact sparse array representation. The trie is constructed on the fly as needed.

  • The param block [cache|naive] tells us whether the in-memory representation is loaded each time a new query is fired (naive), or if we cache previously loaded relations (cache).

  • The param bloc [bfs|dfs] selects the algorithm applied to evaluate the queries. Note that bfs combined with ANY gives a single shortest path, while dfs with ANY gives an arbitrary path.

  • The param dfs is only considered when using ANY|SIMPLE|TRAILS, because ALL|ALL_COUNT don't have an dfs implementation.

  • The params [cache|naive] are only considered when using trie. They don't have any impact when using btree.