A2Cloud

A framework for empowering scientific and high-performance computing via the cloud.

Setup and Dependencies

To get started, simply clone this repository

$ git clone https://github.com/cojomojo/a2cloud.git

To utilize the PERF engine (to extract application features), you must:

• Have Linux perf tools installed
• Have a supported version of Python installed
• Be on an x86_64 architecture (for now)

To utilize the cloud trace engine (to extract cloud instance features), you must:

• Have gfortran installed
• Have superuser privileges on the instance in question

How to Use

To generate a application feature vector, utilize the PERF engine. The PERF engine can be used to extract features of your application on any x86_64 machine, it does not have to be done on the target cloud instances.

To generate a cloud feature vector, utilize the Cloud Trace Engine. This should be done on the cloud instance(s) you are interested in.

The PERF Engine

One way to use the engine is to measure the features of your application:

$ cd a2cloud/perf_engine
$ ./afm.py measure MY_APP ~/codes/my_app arg1 arg2 argN

This will output a JSON file with the raw information. To view this information in a friendly way you can use the PERF engine's report feature:

$ ./afm.py report MY_APP_03122018.json

The Cloud Trace Engine

To use this engine to measure the features of a cloud instance:

$ cd a2clooud/cloud_trace_engine
$ make
$ ./a2cloud-bench.sh # requires superuser

Data Available

The data folder contains application and cloud vectors which have been collected so far. Application vectors can be explored easily using the PERF engine report feature:

$ ./afm.py report a2cloud/data/vectors/applications/**/*.json
===========================================================
../../data/vectors/applications/LULESH/LULESH_10.json
{'spflops': 0.0, 'x87': 16.891292899999996, 'dpflops': 594.24763610000002, 'memory': 40.768000000000001}
arithmetic intensity = 2.7074
===========================================================
../../data/vectors/applications/LULESH/LULESH_30.json
{'spflops': 0.0, 'x87': 3133.6285904000001, 'dpflops': 74272.050185100001, 'memory': 5246.4879999999994}
arithmetic intensity = 2.6915
.
.
.

Read The Paper

The A2Cloud paper is currently under review in IEEE CLOUD 2018.

Contributors

University of the Pacific School of Engineering and Computer Science:

• Cody Balos
• Vivek Pallipuram, Ph.D (primary contact) - vpallipuramkrishnamani@pacific.edu
• Zachariah Abuelhaj
• David De La Vega
• David Mueller, Ph.D
• Chadi El Kari, Ph.D

License

MIT