Memory resource balancing for virtualized computing

Virtualization has become a common abstraction layer in modern data centers. By multiplexing hardware resources into multiple virtual machines (VMs) and thus enabling several operating systems to run on the same physical platform simultaneously, it can effectively reduce power consumption and building size or improve security by isolating VMs. In a virtualized system, memory resource management plays a critical role in achieving high resource utilization and performance. Insufficient memory allocation to a VM will degrade its performance dramatically. On the contrary, over-allocation causes waste of memory resources. Meanwhile, a VM’s memory demand may vary significantly. As a result, effective memory resource management calls for a dynamic memory balancer, which, ideally, can adjust memory allocation in a timely manner for each VM based on their current memory demand and thus achieve the best memory utilization and the optimal overall performance. In order to estimate the memory demand of each VM and to arbitrate possible memory resource contention, a widely proposed approach is to construct an LRU-based miss ratio curve (MRC), which provides not only the current working set size (WSS) but also the correlation between performance and the target memory allocation size. Unfortunately, the cost of constructing an MRC is nontrivial. In this dissertation, we first present a low overhead LRU-based memory demand tracking scheme, which includes three orthogonal optimizations: AVL-based LRU organization, dynamic hot set sizing and intermittent memory tracking. Our evaluation results show that, for the whole SPEC CPU 2006 benchmark suite, after applying the three optimizing techniques, the mean overhead of MRC construction is lowered from 173% to only 2%. Based on current WSS, we then predict its trend in the near future and take different strategies for different prediction results. When there is a sufficient amount of physical memory on the host, it locally balances its memory resource for the VMs. Once the local memory resource is insufficient and the memory pressure is predicted to sustain for a sufficiently long time, a relatively expensive solution, VM live migration, is used to move one or more VMs from the hot host to other host(s). Finally, for transient memory pressure, a remote cache is used to alleviate the temporary performance penalty. Our experimental results show that this design achieves 49% center-wide speedup.

Dynamic properties of wood using the Split-Hopkinson Pressure Bar

Strain rate significantly affects the strength of a material. The Split-Hopkinson Pressure Bar (SHPB) was initially used to study the effects of high strain rate (~103 1/s) testing of metals. Later modifications to the original technique allowed for the study of brittle materials such as ceramics, concrete, and rock. While material properties of wood for static and creep strain rates are readily available, data on the dynamic properties of wood are sparse. Previous work using the SHPB technique with wood has been limited in scope to variability of only a few conditions and tests of the applicability of the SHPB theory on wood have not been performed. Tests were conducted using a large diameter (3.0 inch (75 mm)) SHPB. The strain rate and total strain applied to a specimen are dependent on the striker bar length and velocity at impact. Pulse shapers are used to further modify the strain rate and change the shape of the strain pulse. A series of tests were used to determine test conditions necessary to produce a strain rate, total strain, and pulse shape appropriate for testing wood specimens. Hard maple, consisting of sugar maple (Acer saccharum) and black maple (Acer nigrum), and eastern white pine (Pinus strobus) specimens were used to represent a dense hardwood and a low-density soft wood. Specimens were machined to diameters of 2.5 and 3.0 inches and an assortment of lengths were tested to determine the appropriate specimen dimensions. Longitudinal specimens of 1.5 inch length and radial and tangential specimens of 0.5 inch length were found to be most applicable to SHPB testing. Stress/strain curves were generated from the SHPB data and validated with 6061-T6 aluminum and wood specimens. Stress was indirectly corroborated with gaged aluminum specimens. Specimen strain was assessed with strain gages, digital image analysis, and measurement of residual strain to confirm the strain calculated from SHPB data. The SHPB was found to be a useful tool in accurately assessing the material properties of wood under high strain rates (70 to 340 1/s) and short load durations (70 to 150 μs to compressive failure).

STATE OF CHARGE BALANCING DROOP CONTROL

This thesis presents a load sharing method applied in a distributed micro grid system. The goal of this method is to balance the state-of-charge (SoC) of each parallel connected battery and make it possible to detect the average SoC of the system by measuring bus voltage for all connected modules. In this method the reference voltage for each battery converter is adjusted by adding a proportional SoC factor. Under such setting the battery with a higher SoC will output more power, whereas the one with lower SoC gives out less. Therefore the higher SoC battery will use its energy faster than the lower ones, and eventually the SoC and output power of each battery will converge. And because the reference voltage is related to SoC status, the information of the average SoC in this system could be shared for all modules by measuring bus voltage. The SoC balancing speed is related to the SoC droop factors. This SoC-based load sharing control system is analyzed in feasibility and stability. Simulations in MATLAB/Simulink are presented, which indicate that this control scheme could balance the battery SoCs as predicted. The observation of SoC sharing through bus voltage was validated in both software simulation and hardware experiments. It could be of use to non-communicated distributed power system in load shedding and power planning.