Disk Resource Allocation Based On I/O Access Charactertics Of Virtual Machines

Virtualization makes use of sharing storage to improve the utilization of storage resource and reduce the cost of storage device in cloud environment. But, with the popularity of the I/O intensive applications and the growing number of virtual machines (VMs) sharing storage, the disk I/O contention among VMs is more serious, thus lowering I/O performance of applications in VMs. However, due to the transparency and encapsulation features of virtualization, traditional resource allocation cannot be applied in virtualized environment. Meanwhile, these features of virtualization complicate I/O access of VMs, making the disk I/O allocation more difficult in the virtualized environment. Therefore, disk resource allocation is key to improve I/O performance of virtualization.To end this, the first step is to analyze disk I/O access characteristics in virtualized environment for guiding disk resource allocation. Given that the transparency and encapsulation features of virtualization and sharing hardware impact on the I/O access, using a multi-angle and multi-granularity evaluation analyzes the distribution of addresses of requests from VMs, throughput and latency of VMs. The results show that:(1) the special spatial locality in virtualized environments:regional spatial locality across VMs and sub-regional spatial locality between requests from the same VM;(2) when VMs with different access patters of applications run together or the number of VMs sharing disk resource grows, the interference among VMs increases, leading to the low throughput of VMs and unstable latency of requests, and not ensuring Quality of Service (QoS) of VMs.Then, according to the analysis of special spatial locality in the virtualized environment, a prediction model on I/O access regularity of VMs is present. The model uses a temporal access-density clustering algorithm to predict the access regularity of sub-regional spatial locality of a VM running mixed applications with keeping the transparency of virtualization. Based on the prediction model, a VM spatial-locality-aware disk scheduling framework is designed for exploiting the special spatial locality (i.e., the regional and sub-regional spatial locality) to reduce disk seek and rotational overhead in the virtualized environment. With a spatial-locality-aware heuristic algorithm, the disk scheduling uses the adaptive non-working-conserving mode to schedule I/O requests. The heuristic algorithm takes advantage of the regional spatial locality across VMs and the sub-regional spatial locality prediction from the vNavigator model to decide whether or not to dispatch the request and adjust the waiting time, to guide scheduling I/O requests.Improving disk I/O efficiency is the fundamental way to improve I/O performance of applications in the virtualized environment. The way based on spatial locality to improve bandwidth utilization of VMs does not consider the bandwidth and latency requirements of VMs. On the other hand, the virtualized platform lacks of performance isolation among VMs, resulting in grabbing I/O resource among VMs and being unable to meet the performance requirements of VMs. Therefore, a VM QoS-based disk scheduling, named Flubber, is present to provide QoS guarantees to VMs while maintaining high disk utilization. The scheduler decouples throughput and latency allocation. The high-level throughput control regulates the pending requests from the VMs with an adaptive credit-rate controller, in order to meet the throughput requirements of different VMs and ensure performance isolation. Meanwhile, the low-level latency control, by the virtue of the batch and delay earliest deadline first (BD-EDF) mechanism, manages request queues in block layer, for ensuring the latency requirement of VM while enforcing disk I/O utilization. Considering the regional spatial locality of accesses across VMs, the BD-EDF mechanism batches requests to disk devices without over the deadlines of requests.Last, based on above disk resource allocation, the relationship between I/O performance requirements of virtual machines and disk resource allocation is discussed future. Because of I/O architecture of virtualization, I/O access pattern of VM and characteristics of disk, it is difficult to define the I/O resource requirement of VM. So the disk resource allocation cannot define the suitable parameter on bandwidth allocation. To target the issue, a VM QoS-aware bandwidth parameter model is proposed, to help virtual machine manager to dynamic allocation disk I/O resource and reduce the waste of resource in accordance with the change of I/O access of VM. The model quantizes the change of QoS of VM, then adjust the throughput parameters among VMs in time, to assist virtual machine manager with allocate bandwidth among VMs.In summary, based on the analysis of characteristics of I/O access of VMs, a series of disk I/O allocations are proposed in the virtualized environment to improve disk I/O efficiency and guarantee QoS of VM, so improving the I/O performance of applications in VMs.