HIP based mobility for Cloudlets

Forskningsoutput: AvhandlingMagisteruppsats

Sammanfattning

Computation offloading can be used to leverage the resources of nearby computers to ease the computational burden of mobile devices. Cloudlets are an approach, where the client’s tasks are executed inside a virtual machine (VM) on a nearby computing element, while the client orchestrates the deployment of the VM and the remote execution in it.

Mobile devices tend to move, and while moving between networks, their address is prone to change. Should a user bring their device close to a better performing Cloudlet host, migration of the original Cloudlet VM might also be desired, but their address is then prone to change as well. Communication with Cloudlets relies on the TCP/IP networking stack, which resolves address changes by terminating connections, and this seriously impairs the usefulness of Cloudlets in presence of mobility events.

We surveyed a number of mobility management protocols, and decided to focus on Host Identity Protocol (HIP). We ported an implementation, HIP for Linux (HIPL), to the Android operating system, and assessed its performance by benchmarking throughput and delay for connection recovery during network migration scenarios.
We found that as long as the HIPL hipfw-module, and especially the Local Scope Iden- tifier (LSI) support was not used, the implementation performed adequately in terms of throughput. On the average, the connection recovery delays were tolerable, with an average recovery time of about 8 seconds when roaming between networks. We also found that with highly optimized VM synthesis methods, the recovery time of 8 seconds alone does not make live migration favourable over synthesizing a new VM.

We found HIP to be an adequate protocol to support both client mobility and server migration with Cloudlets. Our survey suggests that HIP avoids some of the limitations found in competing protocols. We also found that the HIPL implementation could benefit from architectural changes, for improving the performance of the LSI support.
Originalspråkengelska
UtgivningsortHelsinki
Förlag
StatusPublicerad - dec 2015
MoE-publikationstypG2 Masteruppsats, polyteknisk masteruppsats

Vetenskapsgrenar

  • 113 Data- och informationsvetenskap

Citera det här