echtzeitsysteme/iflye

Incremental Fast Lightweight (y) virtual network Embedding framework

iflye's Issues

Pipeline VNE algorithm: Fix eMoflon resource folder (../)

Currently, the pipeline-based VNE algorithm (iflye-pipeline) extracts some of the resources to its parent folder, which should be fixed.

Fix minor implementation issues

There are some minor implementation issues left, typically marked with a TODO.

• Pipeline-based algorithms: "Normal" memory metrics are broken.
• CsvUtil error handling. https://github.com/maxkratz/iflye/blob/1bc84da8d29f8ef0b6f89b848bca2d90db743fe0/vne.scenarios/src/scenario/util/CsvUtil.java#L220
• GlobalMetricsManager temporary evaluation fix. https://github.com/maxkratz/iflye/blob/1bc84da8d29f8ef0b6f89b848bca2d90db743fe0/network.metrics/src/metrics/manager/GlobalMetricsManager.java#L190
• Use better data structures in some classes, e.g., use UnifiedMap from the Ecplise Collections in the class IncrementalGurobiSolver for a possible minor performance boost.

Add paths for Windows-based systems

Gurobi update: v9.5.0 to v9.5.1

https://cdn.gurobi.com/wp-content/uploads/2022/03/release-notes-9_5_1.html

Update Gurobi to v9.5.2

The title says everything.

Update Gurobi to v10.0.2

https://www.gurobi.com/downloads/gurobi-optimizer-release-notes-v10-0-2/

Update Gurobi solver

Currently, the framework uses version 8.1.1, build v8.1.1rc0, of the Gurobi solver to make the performance comparable to iDyVE's implementation. However, after I submitted my master thesis, Gurobi can be upgraded to the latest version (9.1.2 at the time of this issue).

Update Apache CSV implementation (v1.11.0+)

There are some occurances of deprecated method usages with Apache commons CSV, e.g. here: https://github.com/Echtzeitsysteme/iflye/blob/main/vne.scenarios/src/scenario/util/CsvUtil.java#L52

TODO: Fix these deprecated method calls

Edit: https://mvnrepository.com/artifact/org.apache.commons/commons-csv

Update Gurobi to v11.0.1

https://support.gurobi.com/hc/en-us/articles/23104396035729-Gurobi-11-0-1-released

`example.gips` does not contain GIPS-based algorithm examples

https://github.com/Echtzeitsysteme/iflye/tree/main/examples/src/examples/gips

Update required Java version to JDK 17

Adapt visualization to master thesis

Currently, the visualization of iflye uses circles as servers and boxes as switches. This is exactly the other way around than in my master thesis.

Rename test project from `test.suite` to `test.suite.iflye`

Rename the test project from test.suite to test.suite.iflye to avoid confusion when having the test.suite.gips project in the same workspace.

Refactor inherited test cases (e.g., VnePmMigration)

Currently, there are various test classes that inherit a super test class and just overwrite the method to create a new algorithm instance to test. As this is a bad coding style, one may refactor the test cases to parametrized tests to remove the high number of classes for each test case.

Example file: https://github.com/maxkratz/iflye/blob/1bc84da8d29f8ef0b6f89b848bca2d90db743fe0/test.suite/src/test/algorithms/pm/migration/VnePmMdvneAlgorithmMigrationTotalCommunicationCostATest.java#L21

PM-Migration VNE algorithm concept extension

As for now, the implementation of the MdVNE algorithm PM with migration enabled tries to re-embed virtual networks until a previously set value of n.

Idea: Try to find a good heuristic to set n to a useful value for a good speed to quality ratio.

Flaky GIPS-based Migration test

The test VnePmMdvneAlgorithmMigrationPathBandwidthBugTest sometimes (rarely) fails, probably because of this bug: Echtzeitsysteme/gips-examples#32

Update Gurobi to v11.0.2

• Release notes: https://www.gurobi.com/documentation/11.0/refman/detailed_release_notes_11.html

Feature request: Rating of matches

Another idea to improve the embedding quality as well as the performance: Rating of matches:

• for embedding candidates
• for removal candidates (in case of migration)

Ideas:

• cost function for matches to calculate their cost/quality

Example criteria:

• remove matches from particularly full or particularly empty substrate servers

The cost function (e.g., with input = all attributes of all nodes of a match) could, for example, be learned by ML approaches.

Add implementation(s) for other ILP solvers like CPLEX

The iDyVE framework also allows the use of other ILP solvers than Gurobi, e.g., IBM CPLEX. Currently, there is no functional implementation for other ILP solvers than Gurobi

Remove all classpath dependencies and move them to Manifest files

Update Gurobi to v10.0.0

https://support.gurobi.com/hc/en-us/articles/360048138771-Gurobi-release-and-support-history

Metric idea: Embedding quality per time

Maybe we can implement a metric that gives a result for "embedding quality per time". This could give a trade-off between long runtimes but good results (pure ILP algorithms) and short runtimes but bad results (pure heuristic algorithm).

Update Apache Commons CLI (v1.70+)

See https://mvnrepository.com/artifact/commons-cli/commons-cli

Fix Windows incompatibility and/or document it

Currently, the framework has some functionality that relies on Linux system calls:

• Memory PID metric (Java implementation uses Linux system calls)
• Bash scripts
• EmoflonGtAppUtils (may feature Linux-style paths, e.g., '/a/b/c')

Update Gurobi to v10.0.3 (+)

Update IBM CPLEX solver to v22.1.0

The title says everything.

Increase test coverage for `ModelFacade` (and other model-related classes)

Currently, the coverage is around ~88%.

Add `gips-ex-to-iflye.sh` script to README.md

https://github.com/Echtzeitsysteme/iflye/blob/main/scripts/gips-ex-to-iflye.sh

Update Gurobi to v11.0.0

https://support.gurobi.com/hc/en-us/articles/360048138771-Gurobi-release-and-support-history

Remove Viatra after releasing new eMoflon-IBeX

Update metamodel after eMoflon data type fix

I've created and described an issue with the comparison of data types with the eMoflon GT engine: eMoflon/emoflon-ibex#408.
After the issue is fixed, adapt all metamodels in this framework.

Alternative: Use type double for all attributes within the metamodel.

Remove custom ILP solver adapter implementation if possible

The implementation of ilp/wrapper/IncrementalIlpSolver.java can potentially be removed due to the fact that emoflon has its own ILP solver interface (at least as I heard!).

Documentation issue: Current need to use feature branches in emoflon-ibex and emoflin-ibex-ui

After merging maxkratz@3b65798 into the branch main, the whole project is only executable while using some feature branches on emoflon-ibex and emoflon-ibex-ui.

• emoflon-ibex must be checked-out to gt-forEach: https://github.com/emoflon/emoflon-ibex/tree/gt-forEach
• emoflon-ibex-ui must be checked-out to forEach_Rule: https://github.com/eMoflon/emoflon-ibex-ui/tree/forEach_Rule

This issue can be closed if both branches mentioned above are merged in the master branches of the corresponding emoflon-ibex* projects.

Remove duplicate vnets.json files

Currently, there are duplicated virtual network JSON files in vne.scenarios/resources/*/vnets.json. Do not forget to also update the evaluation scripts if you remove these files.

Implement removal of substrate switches

Compared to the removal functionality of substrate servers, the framework, i.e., the class ModelFacade.java, should feature a method the remove substrate switches.