ShapeShifter

This repo covers the prototype and tools for ShapeShifter.

Tools for Lambda Enforcer and Kubernetes integration

System prerequisite

For running the tools here, you can directly run them on your Linux server, since we only required the tc command. It is recommended to run in your MEC Kubernetes testbed.

The instruction of our MEC testbed is covered as following:

Kubernetes: with the CNI, WeaveNet.
vRAN, the LTE testbed by NextEPC's core network; and SrsRAN's eNB and UE.

Docker image creation and K8s deployment

Dockerfile and Kubernetes configuration file, ./ss.yml, helps to deploy tools in the space of control plane.

In the K8s configuration file, be careful of the indicated node (nodeSelector) for deployment, the name of image, and the CNI gateway interface (weave in command). You should change them to fit your environment. Also, you can adjust other setup like the running time (default: 1h), preprocessing and post processing, etc.

You can take steps below as reference:

// cd to the root directory of this repo, build the image
$ docker build -t ss .

// deploy shape enforcer tools
$ kubectl create -f ss.yml

Introduction of tools (scripts)

File	Purpose	Input parameter(s)
`create_queue.sh`	create a queue with specific shape rules	interface, dest IP, unique ID of queue, bitrate in mbit, burst size in kbit, latency in msec
`create_single_queue.sh`	same as above, but only expect to create one queue, so no need to specify IP and ID	interface, bitrate in mbit, burst size in kbit, latency in msec
`create_fair_queue.sh`	create a fair queue	interface, dest IP, unique ID of queue
`lc_queue_example.sh`	create single Lc queue with our example parameters	interface
`ls_queue_example.sh`	create single Ls queue with our example parameters	ditto
`reset_interface.sh`	clean the shape rules	ditto
`remove_qdisc.sh`	remove whole qdisc functions on interface, need to attach it again before creating queues, can run `reset_interface.sh` to get it back	ditto
`show_interface.sh`	show the queues on the interface	ditto
`add_delay.sh`/`delete_delay.sh`	add/delete certain number of millisecond delay on a specific network interface	ditto

ShapeShifter requires to shape both ingress and egress traffics. So, we will need to run tool twice for the traffics separately.

As describe in the paper's section VI. IMPLEMENTATION and Fig. 6, we shape the ingress traffic of MEC service at RAN node, and shape its egress traffic on MEC node. While the ingress traffics on RAN node would be several of them, duo to several MEC service, we used to run create_queue.sh to specify each of them one by one; on the other hand, the egress traffic would be set on the vNIC of the MEC server, we can use create_single_queue.sh, since that vNIC is dedicated for this application.

Here is the example for calling the tool:

// show the name of ShapeShifter tool in control plane
$ kubectl get pod -n kube-system
NAME                             READY   STATUS    RESTARTS       AGE
:
ss-mec-5lcb2                     1/1     Running   0              3m
ss-vran-x68ms                    1/1     Running   0              3m
:

// apply ingress traffic of MEC service on RAN
$ kubectl exec -it -n kube-system ss-vran-x68ms -- sh ./create_queue.sh weave 10.32.0.2 1 1 10 1

// apply egress traffic by the Lc example 
$ kubectl exec -it -n kube-system ss-mec-5lcb2 -- sh ./lc_queue_example.sh vethwepl31c95c2

MEC application usecases in paper

Streamer: The latency-sensitive (Ls) use case, tagged as ML.
ROS: The latency-critical (Lc) use case, tagged as IA.

Paper reference

@inproceedings{shapeshifter,
  author    = {Valentin Rakavic
               Ke{-}Jou Hsu and
               Ketan Bhardwaj and
               Ada Gavrilovska and
               Liljana Gavrilovska},
  title     = {{ShapeShifter: Resolving the Hidden Latency Contention Problem in MEC}},
  booktitle = {Proceedings of the 7th {ACM/IEEE} Symposium on Edge Computing, {SEC}
               2022, Bellevue, WA, USA, December},
  publisher = {{ACM/IEEE}},
  year      = {2022},
  url       = {https://doi.org/10.1109/SEC54971.2022.00026},
  doi       = {10.1109/SEC54971.2022.00026}
}

GTkernel / ShapeShifter