mycloud 0.51 (experimental)

Work distribution for small clusters.

MyCloud

Leverage small clusters of machines to increase your productivity.

MyCloud requires no prior setup; if you can SSH to your machines, then it will work out of the box. MyCloud currently exports a simple mapreduce API with several common input formats; adding support for your own is easy as well.

Usage

Starting your cluster:

import mycloud

cluster = mycloud.Cluster(['machine1', 'machine2'])

# or use defaults from ~/.config/mycloud
# cluster = mycloud.Cluster()

Map over a list:

result = cluster.map(compute_factors, range(1000))

ClientFS makes accessing local files seamless!

def my_worker(filename):
  do_work(mycloud.fs.FS.open(filename, 'r'))

cluster.map(['client:///my/local/file'], my_worker)

Use the MapReduce interface to easily handle processing of larger datasets:

from mycloud.mapreduce import MapReduce, group
from mycloud.resource import CSV
input_desc = [CSV('client:///path/to/my_input_%d.csv') % i for i in range(100)]
output_desc = [CSV('client:///path/to/my_output_file.csv')]

def map_identity(kv_iter, output):
  for k, v in kv_iter:
    output(k, int(v[0]))

def reduce_sum(kv_iter, output):
  for k, values in group(kv_iter):
    output(k, sum(values))

mr = MapReduce(cluster, map_identity, reduce_sum, input_desc, output_desc)

result = mr.run()

for k, v in result[0].reader():
  print k, v

Performance

It is, keep in mind, written entirely in Python.

Some simple operations I've used it for (6 machines, 96 cores):

• Sorting a billion numbers: ~5m
• Preprocessing 1.3 million images (resizing and SIFT feature extraction): ~1 hour

Input formats

Mycloud has builtin support for processing the following file types:

• LevelDB
• CSV
• Text (lines)
• Zip

Adding support for your own is simple - just write a resource class describing how to get a reader and writer. (see resource.py for details).

Why?!?

Sometimes you're developing something in Python (because that's what you do), and you decide you'd like it to be parallelized. Our current options are multiprocessing (limiting us to a single machine) and Hadoop streaming (limiting us to strings and Hadoop's input formats).

Also, because I could.

Credits

MyCloud builds on the phenomonally useful cloud serialization, SSH/Paramiko, and LevelDB libraries.