ESI Talents

Sujet de Stage N°: 299

Sujet de Stage
Validation

Sujet de stage N° 299

Type*

Specialites*

Organisme*

Coencadrants*

Promoteur*

BAGAA Miloud

Email promoteur*

Tel promoteur

Intitule*

Efficient and Fast Service Relocation Accros Multiple Edges

Resume*

The recent advances and the progressive digital transformation, and shifting toward network and clouds/edges create high disruptive forces shaking up the network era. From another side, the emerging verticals, such as Ultra-Reliable Low-Latency Communication (URLLC), require aggressive and conflicting requirements. These verticals target close to zero time latency, a high data rate with ten times more than the current network traffic, five-nines network availability, as well as a more dense network with 10 - 100 more connected users and devices. For this reason, there is a need to host services at the multi-access edge computing (MEC) in the proximity of the users. However, as users move from a location to another that enforces the system to move also his services enable the follow me edge concept. For this reason, we need to ensure service continuity while following user mobility. To support the migration of stateful services, efficient follow me concepts should be designed and implemented to provide the service relocation while minimizing the time of service distribution (i.e., downtime). The service migration can be enabled using either CRIU tool or Distributed Shared Memory (DSM) mechanism.

The roadmap to this topic includes an evaluation of the current performance limitations of CRIU Page Server and kernel-based software DSM system to enable the migration of applications among nodes on MECs. The objective is to reduce the migration time (i.e., downtime) to be close to zero. To achieve the desired goals, the students will conduct their investigation through the development of migration prototypes that leverage both CRIU and DSM systems.

The students will first leverage CRIU internals to implement the copy of data from/into the Linux kernel, while focusing primarily on techniques to parallelize the transference of memory pages (dumps) and restore procedure. Currently, CRIU allows the user to do multiple rounds of partial memory dumps, also known as pre-dumps, while the container/application is still running. However, to enable the migration, the container/application must be frozen to allow the final dump. This approach harms the service performance and readiness as it introduces longer downtimes. The students will enhance and use the migration process by leveraging ideas related to CRIU, such as Lazy Migration and Page Server. They will investigate the utilization of network support subsystems, including RabbitMQ, ZeroMQ, and NGINX, and the proposal of a pipelined method that enables partial restore system calls. The latter will be leveraged for enabling pre-dumps to pre-populate the data into the kernel.

Finally, the students will conduct their investigation through the development of migration systems prototypes that leverage on DSM, such as popcorn. The latter will be harnessed for replicating the user-level memory pages on a remote host and subsequently reconfiguration of the process virtual memory address space to use the remote memory as the master/primary memory pages. Moreover, the students will investigate how to create a new process infrastructure using the Linux kernel PID namespace facilities on a second host using the user-level memory pages on the original host to populate the new processes virtual address space. The expected results of this investigation are an in-depth analysis of how an in-kernel DSM subsystem can be leveraged to build a new execution environment with migration support, allowing a smooth migration process including memory pages and network connections.

While the main research would be carried out at ESI, the student will be advised and guided physically and remotely via virtual meetings that include his local advisor at ESI and experts from Aalto University Moas!c Lab (http://www.mosaic-lab.org/) and Federal University of Rio de Janeiro Compass Laboratory (https://www.compasso.ufrj.br/).

Objectifs*

1. Etude and study about CRIU tool to migrate services across different clouds. 
2. Investigation about different tools and concepts, such as Lazy Migration and Page Server, for reducing the downtime.
3. Etude and study about existing tools related with DSM system, such as popcorn.
4. Leverage the DSM tools for creating new process infrastructure that enables the replication and synchronization of the process.  
5. Evaluate performances of all the proposed tool and solutions.

Resultats attendus*

1. Etude and study about CRIU tool to migrate services across different clouds. 
2. Investigation about different tools and concepts, such as Lazy Migration and Page Server, for reducing the downtime.
3. Etude and study about existing tools related with DSM system, such as popcorn.
4. Leverage the DSM tools for creating new process infrastructure that enables the replication and synchronization of the process.  
5. Evaluate performances of all the proposed tool and solutions.

Antecedents

Moyens informatiques

Echeancier*

Step 1
Study the state of the art about Process in Linux, CRIU and DSM. 
02 months
Step 2
Study and investigation about different alternatives to provides a smooth service migration process by leveraging DSM concept and CRIU tool.
03 months
Step 3
Conception and development of the desired framework. 
05 months
Step 4
Thesis writing.
01 month

Bibliographie

Statut validation*

Reponse aux experts

Reserve pour*

Projet recherche

Notification

Avis sur le Sujet de Stage

Intitule	Organisme	Debut	Fin	Avis	Commentaire
Efficient and Fast Service Relocation Accros Multiple Edges	MOSAIC : MOSA!C Lab, Aalto University	09/09/2020	09/30/2020	Avis favorable	—
Efficient and Fast Service Relocation Accros Multiple Edges	MOSAIC : MOSA!C Lab, Aalto University	09/09/2020	09/22/2020	Avis favorable	—
Efficient and Fast Service Relocation Accros Multiple Edges	MOSAIC : MOSA!C Lab, Aalto University	09/09/2020	09/30/2020	Avis favorable	—