| Age | Commit message (Collapse) | Author |
|---|
|
This change adds support for virtual machines and hypervisors to the
_opendc-compute_ module. Moreover, this change also includes VM trace
reading capabilities.
|
|
These experiments were originally designed for the SC18 paper.
|
|
This change remodels our model for workloads and compute resources in
order to support VM/container functionality where multiple workloads run
on a single machine.
In particular, we make the following changes:
- Move the compute-related source code into the `opendc-compute` module.
- Change from application-based model to image-based model, where the
image has exclusive access over the machine, instead of applications
that share the machine. We may model in the future again
applications/operating system, but at the moment, we do not need this
granularity, given that the workload traces can be simulated using VMs.
|
|
|
|
This change adds a prototype implementation of the revised version of
the API of version 2.0 of the simulator.
|
|
This change introduces the revised API design for version 2.0 of the
OpenDC simulator. This version drops built-in support for Java and
instead opts to build on Kotlin coroutines to simplify the API surface.
During development of and experimentation with the previous API for version
2.x, we found that the design based on Akka Typed was too limiting and caused
too much boilerplate for the models we needed to implement. Essential
patterns such as request-response were found to be hard to implement
with only a single mailbox. Moveover, limiting each actor's mailbox to a
single type hindered composition and often resulted in unchecked casts
or the type being changed to `Any`, eliminating the type-safety of the
API.
In this revised API design, a simulation is now represented as the interplay of
logical processes that communicate via multiple message passing channels.
We use Kotlin coroutines to describe the behavior of the processes.
The API has been design from the start to take into account
distributed/parallel simulations by disallowing messages from arbitrary
processes, which was possible in the previous design. Instead, the
'communication graph' is known during runtime as procsses must register
themselves before being able to send/receive messages to/from channels.
We are still figuring out process/channel identity and supervision. Currently,
all logical processes run on a single level, instead of being
hierachical. However, this might change in the future.
|
|
This change renames the main module of the odcsim library to odcsim-api,
since it mainly contains the interfaces to be used by consumers of the
API and implemented by the various frameworks.
|
|
|
|
This change adds the initial version of the port of the OpenDC
simulation model to version 2.x of the simulator.
The simulation model has been reworked to support immutability and
event-driven simulation, with speed-ups up to 75x.
|
|
This change adds a testkit for synchronously testking Behavior
implementations.
|
|
This change migrates the build configuration for Gradle in Groovy to
Kotlin, where possible. The scripts in the `gradle/` directory have not
been migrated to extensive use of dynamicism.
|
