From e9a29f5440a7c43b7dbb9cc447d047b51f53e9ac Mon Sep 17 00:00:00 2001 From: Rick Harris Date: Tue, 17 May 2011 10:45:19 -0500 Subject: [PATCH] First cut at least cost scheduler --- nova/scheduler/least_cost.py | 79 +++++++++++++++++++++++++ nova/tests/test_least_cost_scheduler.py | 39 ++++++++++++ 2 files changed, 118 insertions(+) create mode 100644 nova/scheduler/least_cost.py create mode 100644 nova/tests/test_least_cost_scheduler.py diff --git a/nova/scheduler/least_cost.py b/nova/scheduler/least_cost.py new file mode 100644 index 000000000..75dde81ca --- /dev/null +++ b/nova/scheduler/least_cost.py @@ -0,0 +1,79 @@ +import collections + +# TODO(sirp): this should be just `zone_aware` to match naming scheme +# TODO(sirp): perhaps all zone-aware stuff should go under a `zone_aware` +# module +from nova.scheduler import zone_aware_scheduler + +class LeastCostScheduler(zone_aware_scheduler.ZoneAwareScheduler): + def get_cost_fns(self): + """Returns a list of tuples containing weights and cost functions to + use for weighing hosts + """ + cost_fns = [] + + return cost_fns + + def weigh_hosts(self, num, specs, hosts): + """ + Returns a list of dictionaries of form: + [ {weight: weight, hostname: hostname} ] + """ + # FIXME(sirp): weigh_hosts should handle more than just instances + cost_fns = [] + hosts = [] + cost_hosts = weighted_sum(domain=hosts, weighted_fns=self.get_cost_fns()) + + # TODO convert hosts back to hostnames + weight_hostnames = [] + return weight_hostnames + +def normalize_list(L): + """Normalize an array of numbers such that each element satisfies: + 0 <= e <= 1 + """ + if not L: + return L + max_ = max(L) + if max_ > 0: + return [(float(e) / max_) for e in L] + return L + +def weighted_sum(domain, weighted_fns, normalize=True): + """ + Use the weighted-sum method to compute a score for an array of objects. + Normalize the results of the objective-functions so that the weights are + meaningful regardless of objective-function's range. + + domain - input to be scored + weighted_fns - list of weights and functions like: + [(weight, objective-functions)] + + Returns an unsorted list like: [(score, elem)] + """ + # Table of form: + # { domain1: [score1, score2, ..., scoreM] + # ... + # domainN: [score1, score2, ..., scoreM] } + score_table = collections.defaultdict(list) + + for weight, fn in weighted_fns: + scores = [fn(elem) for elem in domain] + + if normalize: + norm_scores = normalize_list(scores) + else: + norm_scores = scores + + for idx, score in enumerate(norm_scores): + weighted_score = score * weight + score_table[idx].append(weighted_score) + + # Sum rows in table to compute score for each element in domain + domain_scores = [] + for idx in sorted(score_table): + elem_score = sum(score_table[idx]) + elem = domain[idx] + domain_scores.append(elem_score) + + return domain_scores diff --git a/nova/tests/test_least_cost_scheduler.py b/nova/tests/test_least_cost_scheduler.py new file mode 100644 index 000000000..a3a18a09f --- /dev/null +++ b/nova/tests/test_least_cost_scheduler.py @@ -0,0 +1,39 @@ +from nova import test +from nova.scheduler import least_cost + +MB = 1024 * 1024 + +class FakeHost(object): + def __init__(self, host_id, free_ram, io): + self.id = host_id + self.free_ram = free_ram + self.io = io + +class WeightedSumTest(test.TestCase): + def test_empty_domain(self): + domain = [] + weighted_fns = [] + result = least_cost.weighted_sum(domain, weighted_fns) + expected = [] + self.assertEqual(expected, result) + + def test_basic_costing(self): + hosts = [ + FakeHost(1, 512 * MB, 100), + FakeHost(2, 256 * MB, 400), + FakeHost(3, 512 * MB, 100) + ] + + weighted_fns = [ + (1, lambda h: h.free_ram), # Fill-first, free_ram is a *cost* + (2, lambda h: h.io), # Avoid high I/O + ] + + costs = least_cost.weighted_sum(domain=hosts, weighted_fns=weighted_fns) + + # Each 256 MB unit of free-ram contributes 0.5 points by way of: + # cost = weight * (score/max_score) = 1 * (256/512) = 0.5 + # Each 100 iops of IO adds 0.5 points by way of: + # cost = 2 * (100/400) = 2 * 0.25 = 0.5 + expected = [1.5, 2.5, 1.5] + self.assertEqual(expected, costs)