Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
smastaplus
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Model registry
Operate
Environments
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
GitLab community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
Maxime MORGE
smastaplus
Commits
4c2a162c
Commit
4c2a162c
authored
2 years ago
by
Maxime MORGE
Browse files
Options
Downloads
Patches
Plain Diff
PSSI: README.md of XP
parent
a62c22f8
No related branches found
No related tags found
No related merge requests found
Changes
1
Show whitespace changes
Inline
Side-by-side
Showing
1 changed file
doc/experiments/README.md
+10
-11
10 additions, 11 deletions
doc/experiments/README.md
with
10 additions
and
11 deletions
doc/experiments/README.md
+
10
−
11
View file @
4c2a162c
...
...
@@ -17,19 +17,19 @@ The following features are evaluated:
-
[
Comparison with Distributed Constraint Optimization Problem (DCOP) solver
](
#comparison-with-distributed-constraint-optimization-problem-dcop-solver
)
-
[
Swap
](
#swap
)
We consider the simulated cost c^S(t,n) as the effective cost of performing the task t
We consider the simulated cost
$
c^S(t,n)
$
as the effective cost of performing the task t
by the node n. :
-
with a perfect knowledge of the computing environment, c^S(t,n) = c(t,n);
-
with half of the nodes slowing down, c^S(t,n) = 2
*
c(t,n) if n mod 2=1 and c^S(t,n) = c(t,n) otherwise.
-
with a perfect knowledge of the computing environment,
$
c^S(t,n) = c
ost
(t,n)
$
;
-
with half of the nodes slowing down,
$
c^S(t,n) = 2
*
c(t,n)
$
if
$
n mod 2=1
$
and
$
c^S(t,n) = c
ost
(t,n)
$
otherwise.
This is the reason why e distinguish the two following metrics:
-
the simulated flowtime C^S(A) for an allocation A according to the simulated costs;
-
the real flowtime C^(At) for an allocation A according to the task completion times which are measured.
-
the simulated flowtime
$
C^S(A)
$
for an allocation A according to the simulated costs;
-
the real flowtime
$
C^(At)
$
for an allocation A according to the task completion times which are measured.
The performance improvement rate is defined:
γ = (C^R(A
O
) - C^R(Ae))/C^R(A0)
$
γ = (C^R(A
0
) - C^R(Ae))/C^R(A0)
$
where
Ae
is the allocation where tasks are executed and
A0
is the initial allocation.
where
$A_e$
is the allocation where tasks are executed and
$A_0$
is the initial allocation.
During the consumption of the initial allocation, we consider the following metrics :
-
the flowtime of the current allocation according to the completion times of tasks
...
...
@@ -108,7 +108,7 @@ time scale) of 128 tasks. It confirms that the best flowtime
achieved is that obtained by our strategy when it is run concurrently with the
executed concurrently with the consumption process. The responsiveness of our strategy is explained by
the reallocation of tasks bundles that represent almost half of them thanks to multiple
concurrent bilateral negotiations that begin even before the consumption process (after
$
0.1
$
second).
concurrent bilateral negotiations that begin even before the consumption process (after 0.1 second).

...
...
@@ -144,11 +144,10 @@ flowtime of the initial random allocation. Even if the simulated flowtime of
the SSI-based allocation is very good, this method
method delays the consumption process in order to first obtain a balanced
allocation and therefore penalises the achieved flowtime. Conversely, the flowtime achieved
by the reallocation methods
reallocation methods
applied during the consumption process,
by the reallocation methods applied during the consumption process,
whether this is our strategy or whether it is based on PSI auctions, is better than the achieved
flowtime realised from random initial allocation. Furthermore, the
flowtime of the reallocation methods is bounded by the
the simulated flowtime of the reallocation (if an oracle calculates the
flowtime of the reallocation methods is bounded by the simulated flowtime of the reallocation (if an oracle calculates the
reallocation in constant time). These methods improve the flowtime
by reallocating, during the consumption process, non-local tasks whose delegation reduces the cost.
As the number of delegations in strategy is greater than the number of delegations in the PSI-based
...
...
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment