D-Track: Efficient Fake Dirty Page Tracker for Live VM Migration

Abstract

Abstract Cloud computing offers efficient and reliable services to millions of users worldwide, ensuring uninterrupted service. They use virtualization to offer services by hosting multiple virtual machines (VMs) on a single physical server. To maintain service continuity during hardware failures or server maintenance, live virtual machine migration techniques transfer virtual machines to different physical servers without interrupting service. It is challenging to accurately predict the total migration time in Pre-Copy migration due to the unpredictable nature of convergence. The iteration count required to transfer memory pages cannot be precisely estimated due to varying dirty page rates and the highly dynamic nature of network bandwidth. In extreme cases, rapid dirty page generation and low network bandwidth can lead to a prolonged total migration time or even failure in migration. Therefore, rapid dirty page generation is a significant problem in Pre-Copy migration. A key issue in this process is generating fake dirty pages, mainly caused by silent store instructions and defects in dirty page tracking. Previous solutions, such as SHA1-based fake dirty prevention algorithms, did not account for the overhead associated with hash computation on migration. This paper presents D-Track, an efficient fake dirty page tracking mechanism that mitigates redundant page transfers by continuously tracking dirty pages during migration. D-Track reduces the total migration time by 20% - 40% when combined with compression techniques while maintaining acceptable application performance degradation. Evaluations of memory-intensive, CPU-intensive, and multiple-intensive workloads demonstrate its effectiveness in improving migration efficiency without significant degradation in VM performance.