Within SITIO project we developed several tools for calculating and predicting costs of running IT systems on various clouds. Recently we open-sourced one of such tools – python library ‘cloud-calculator’. Source is here: https://github.com/HEP-KBFI/ccc. Using easy_tools it can be installed as easy as “easy_install cloud-calculator”.
In this blob post we show sample usage of this library. We start by getting usage statistics. In general, statistics can be either simulated or acquired from one of the monitoring solutions (e.g. Ganglia, Zabbix or collectd). Let’s generate usage statistics for a 12-month period. Result of a ‘generate_random_usage’ function is a list of accumulated monthly usage. The last parameter – ‘spikes’, ‘flat’ or ‘semi-flat’ – is a randomization parameter, which describes variation of the usage.
from sitio.common.utils import generate_random_usage
# network (GB/month) and storage (avg GB/month)
storage_used = generate_random_usage(40, 'spikes')
network_used_in = generate_random_usage(2, 'spikes')
network_used_out = generate_random_usage(20, 'spikes')
# consumed VM time (h/month) and used memory (GB/h*month), normalized to AWS CPU units
cpu_usage = generate_random_usage(5000, ‘spikes’)
mem_usage = generate_random_usage(12000, ‘spikes’)
Now that we have some stats, let's perform basic analysis:
from sitio.analyser import aws, rackspace
# Calculate storage cost on two clouds: AWS and Rackspace
# Pricelists in csv format are located in sitio/analyser/pricelist folder.
ebs_storage_cost, s3_storage_cost = aws.get_storage_costs(storage_used)
rack_storage_costs = rackspace.get_storage_costs(storage_used)
print "Storage costs on AWS: $%s, $%s"
print "Storage costs on Rackspace: $%s" % rack_storage_costs
To calculate migration costs in a straightforward manner - simply migrating all the data - we need to know two things: cost of moving out and cost of moving in.
aws_to_rack_migration_cost = aws.get_network_out_price(storage_used) + \
print "Cost of migrating data from AWS to Rackspace: $%s " % aws_to_rack_migration_cost
To get an estimate of how much a
certain computational load would cost on a cloud, we provide an implementation of the method described in ''Towards a model for cloud computing cost estimation with reserved resources'', CLOUDCOMP'2010. It works by finding cheapest fit using LP of VM time and RAM consumption to cloud provider offering. The last boolean argument of the function signifies whether precise solution is sought (True, results in solving NP problem) or approximate is enough (False, fast).
used_reserved_instances, ec2_cost = aws.get_optimal_ec2(cpu_usage, mem_usage, False)
print "Approximate cost on AWS EC2: $%s" % ec2_cost
To use the same cpu_usage for Rackspace, we need to normalize use by Rackspace benchmark. This is an open issue, but some ad hoc testing showed that it is approximately x5 EC2 CU. YMMV.
cpu_usage = [5*x for x in cpu_usage]
rackspace_cost = rackspace.get_optimal_rackspace(cpu_usage, mem_usage, False)
print "Approximate cost on Rackspace: $%s" % rackspace_cost