FlashX tools

Interact with SAFS

utils/SAFS-util is a tool provided by SAFS that helps users to interact with SAFS. It provides a set of commands to access SAFS.

SAFS-util conf_file command ...
The supported commands are
	create file_name size: create a file with the specified size
	delete file_name: delete a file
	help: print the help info
	list: list existing files in SAFS
	load file_name [ext_file]: load data to the file
	load_part file_name ext_file part_id: load part of the file to SAFS
	verify file_name [ext_file]: verify data in the file
	export file_name ext_file: export an SAFS file to Linux filesystem
	info file_name: show the information of an SAFS file
	rename file_name new_name: rename an SAFS file

The most useful commands are:

• create: create an SAFS file with the specified size
• delete: delete an SAFS file
• list: list all files in SAFS
• load: load data to a SAFS file named file_name. Users can load data from a Linux file to the SAFS file by providing a second argument. If a Linux file isn't provided, the SAFS file is filled with 0.
• export: export an SAFS file to Linux filesystem
• info file_name: show the metadata information of an SAFS file
• rename file_name new_name: rename an SAFS file

Construct FlashGraph graphs

el2fg takes a file of edge lists in the text format and converts them to the FlashGraph format, which is adjacency list in the binary format. The edge list file(s) can be gzip'd. The gzip'd file needs to have a filename extension .gz in order to be considered as a gzip'd file. el2fg generates two binary files: a graph file that contains the adjacency lists of the graph and an index file that contains the locations of vertices in the graph file. By default, el2fg keeps all intermediate data in memory.

matrix/utils/el2fg [options] conf_file edge_file graph_name
-u: undirected graph
-U: unqiue edges
-e: use external memory
-s size: sort buffer size
-g size: groupby buffer size
-t type: the edge attribute type

el2fg takes a FlashX configuration file, a text edge list file and a graph name, and outputs two files: graph_name.adj and graph_name.index.

• -u: To construct an undirected graph, users have to explicitly enable this flag.
• -U: enable this flag to remove redundant edges in a graph.
• -e: enable this flag to use disks to construct a graph. It requires users to configure SAFS correctly.
• -t type: users can specify the edge attribute type. When this option is specified, the input edge list has to have three columns and the third column provides the edge attribute. Currently, four attribute types are supported. "I": 32-bit integer attributes, "L": 64-bit integer attributes, "F": single-precision float-point, "D": double-precision float-point.

An example:

To convert a directed graph in the edge list format to the FlashGraph format, we can run the following command:

flash-graph/tools/el2al -w wiki-Vote.adj wiki-Vote.index wiki-Vote.txt

This converts an edge list in wiki-Vote.txt to the FlashGraph format and generate two files: wiki-Vote.adj-v4 and wiki-Vote.index-v4. wiki-Vote.adj-v4 contains the graph data and wiki-Vote.index-v4 contains the index to the graph data.

To convert an undirected graph in the edge list format to the FlashGraph format, we run:

flash-graph/tools/el2al -w -u facebook.adj facebook.index facebook_combined.txt

When converting a large graph, el2al uses a lot of memory if it keeps all intermediate data in memory, and takes a long time to convert a graph. el2al can keep intermediate data on disks and run in parallel. When -d flag is used, el2al uses stxxl to store intermediate data on disks. Note: users may need to create .stxxl in the current directory to configure the stxxl library where running el2al with -d flag. The page describes how to configure stxxl. -T flag specifies the number of threads used for graph format conversion.

For example, the following command converts a twitter graph with around 60 million vertices in the text edge list format to the adjacency list format out of core and in parallel.

el2al -w -T 32 -d twitter.adj twitter.index twitter_rv.net.gz

Construct FlashMatrix sparse matrices