GTkernel / FAM-Graph

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

FAM-Graph

FAM-Graph is a semi-external graph processing system designed for the disaggregated FAM (Fabric Attached Memory) architecture. Please contact Daniel Zahka (dzahka3@gatech.edu) for access to the system or other questions.

Build Instructions

FAM-Graph uses cmake

mkdir build
cd build
cmake ..
make

Configuring you systems for RDMA

Increasing Memory Limit

It is important to increase the amount of memory that you are allowed to pin, using ulimit.

Configure IB interfaces with IP addresses for IPoIB

FAM-Graph uses the RDMA connection manager which in turn uses IPoIB to bootstrap QP connections. We have configured our client interface with:

sudo ifconfig ib0 192.168.12.1 netmask 255.255.255.0 up

And our server with:

sudo ifconfig ib0 192.168.12.3 netmask 255.255.255.0 up

Running the system

The following examples use the input MOLIERE_2016 (from the next section). Here are the commands run on both the client and server.

Breadth First Search

Server Command

./main -m server -e /mnt/graph1/fam-graph/MOLIERE.adj -t 10

Client Command

./main -m client -k bfs -i /mnt/graphs/fam-graph/MOLIERE.adj -t 10 --start-vertex 1

Pagerank

Server Command

./main -m server -e /mnt/graph1/fam-graph/twitter7.adj -t 10

Client Command

./main -m client -k pagerank_delta -i /mnt/graphs/fam-graph/twitter7.idx -t 10

Weakly Connected Components

Server Command

./main -m server -e /mnt/graph1/fam-graph/MOLIERE.adj -t 10

Client Command

./main -m client -k CC -i /mnt/graphs/fam-graph/MOLIERE.idx -t 10

K-Core Decomposition

Server Command

./main -m server -e /mnt/graph1/fam-graph/MOLIERE.adj -t 10

Client Command

./main -m client -k kcore --kcore-k 200 -i /mnt/graphs/fam-graph/MOLIERE.idx -t 10

Maximal Independent Set

Server Command

./main -m server -e /mnt/graph1/fam-graph/MOLIERE.adj -t 10

Client Command

./main -m client -k MIS -i /mnt/graphs/fam-graph/MOLIERE.idx -t 10

Obtaining Inputs

The inputs used in the FAM-Graph paper are from https://sparse.tamu.edu/. The exact inputs used are:

About