HeteroVisor

Abstract

This paper presents HeteroVisor, a heterogeneity-aware hypervisor, that exploits resource heterogeneity to enhance the elasticity of cloud systems. Introducing the notion of 'elasticity' (E) states, HeteroVisor permits applications to manage their changes in resource requirements as state transitions that implicitly move their execution among heterogeneous platform components. Masking the details of platform heterogeneity from virtual machines, the E-state abstraction allows applications to adapt their resource usage in a finegrained manner via VM-specific 'elasticity drivers' encoding VM-desired policies. The approach is explored for the heterogeneous processor and memory subsystems evolving for modern server platforms, leading to mechanisms that can manage these heterogeneous resources dynamically and as required by the different VMs being run. HeteroVisor is implemented for the Xen hypervisor, with mechanisms that go beyond core scaling to also deal with memory resources, via the online detection of hot memory pages and transparent page migration. Evaluation on an emulated heterogeneous platform uses workload traces from real-world data, demonstrating the ability to provide high on-demand performance while also reducing resource usage for these workloads.

Nimble Page Management for Tiered Memory Systems

“…Additionally, to obtain page access information autoNUMA offlines pages for profiling, and such offlining causes unpredictable memory access latency and bandwidth. These two issues have spurred a range of academic work on two-level memories; however, prior studies have generally focused on page migration policy while assuming page migration mechanisms should be sufficient [2,9,15,16,24,39,46,59,69,82].…”

MEMTIS: Efficient Memory Tiering with Dynamic Page Classification and Page Size Determination

“…Several systems [14,32,49,76,78] use page faults to track memory access. Others [22,27,30,48,60,84] check whether a page is accessed by scanning the associated reference bit; for mmap-ed pages, the processor turns on the reference bit of a page, and for file-backed pages, the OS updates this bit. The page table-based approaches have several critical limitations.…”

“…Software-based tiered memory system. Prior softwarecontrolled tiered memory systems have explored the design space in various aspects including page migration [34,71,83,84], hotness detection [4,15,41], page replacement [28,48], and kernel object tiering [31,36] at different layers such as the application [29,44,69,79], library [21,54,68], and OS [14,27,30,32,76,78]. HotBox [14] suggests not using huge pages in tiered memory systems due to the hotness fragmentation in huge pages.…”

“…Limitations of the state-of-the-art systems. Unfortunately, existing works in tiered memory systems [14,21,27,29,30,32,44,48,49,54,68,69,76,78,79,84] and huge page management approaches [23,26,38,43,47,50,53,57,58,67,87] fail to meet the above criteria.…”