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.

Analysis of in-cylinder pressure transducer data quality utilizing a SIDI turbocharged engine

In-cylinder pressure transducers have been used for decades to record combustion pressure inside a running engine. However, due to the extreme operating environment, transducer design and installation must be considered in order to minimize measurement error. One such error is caused by thermal shock, where the pressure transducer experiences a high heat flux that can distort the pressure transducer diaphragm and also change the crystal sensitivity. This research focused on investigating the effects of thermal shock on in-cylinder pressure transducer data quality using a 2.0L, four-cylinder, spark-ignited, direct-injected, turbo-charged GM engine. Cylinder four was modified with five ports to accommodate pressure transducers of different manufacturers. They included an AVL GH14D, an AVL GH15D, a Kistler 6125C, and a Kistler 6054AR. The GH14D, GH15D, and 6054AR were M5 size transducers. The 6125C was a larger, 6.2mm transducer. Note that both of the AVL pressure transducers utilized a PH03 flame arrestor. Sweeps of ignition timing (spark sweep), engine speed, and engine load were performed to study the effects of thermal shock on each pressure transducer. The project consisted of two distinct phases which included experimental engine testing as well as simulation using a commercially available software package. A comparison was performed to characterize the quality of the data between the actual cylinder pressure and the simulated results. This comparison was valuable because the simulation results did not include thermal shock effects. All three sets of tests showed the peak cylinder pressure was basically unaffected by thermal shock. Comparison of the experimental data with the simulated results showed very good correlation. The spark sweep was performed at 1300 RPM and 3.3 bar NMEP and showed that the differences between the simulated results (no thermal shock) and the experimental data for the indicated mean effective pressure (IMEP) and the pumping mean effective pressure (PMEP) were significantly less than the published accuracies. All transducers had an IMEP percent difference less than 0.038% and less than 0.32% for PMEP. Kistler and AVL publish that the accuracy of their pressure transducers are within plus or minus 1% for the IMEP (AVL 2011; Kistler 2011). In addition, the difference in average exhaust absolute pressure between the simulated results and experimental data was the greatest for the two Kistler pressure transducers. The location and lack of flame arrestor are believed to be the cause of the increased error. For the engine speed sweep, the torque output was held constant at 203 Nm (150 ft-lbf) from 1500 to 4000 RPM. The difference in IMEP was less than 0.01% and the PMEP was less than 1%, except for the AVL GH14D which was 5% and the AVL GH15DK which was 2.25%. A noticeable error in PMEP appeared as the load increased during the engine speed sweeps, as expected. The load sweep was conducted at 2000 RPM over a range of NMEP from 1.1 to 14 bar. The difference in IMEP values were less 0.08% while the PMEP values were below 1% except for the AVL GH14D which was 1.8% and the AVL GH15DK which was at 1.25%. In-cylinder pressure transducer data quality was effectively analyzed using a combination of experimental data and simulation results. Several criteria can be used to investigate the impact of thermal shock on data quality as well as determine the best location and thermal protection for various transducers.

MODELING AND PROTOCOL DESIGN FOR EMERGING WIRELESS SYSTEMS

Mobile Mesh Network based In-Transit Visibility (MMN-ITV) system facilitates global real-time tracking capability for the logistics system. In-transit containers form a multi-hop mesh network to forward the tracking information to the nearby sinks, which further deliver the information to the remote control center via satellite. The fundamental challenge to the MMN-ITV system is the energy constraint of the battery-operated containers. Coupled with the unique mobility pattern, cross-MMN behavior, and the large-spanned area, it is necessary to investigate the energy-efficient communication of the MMN-ITV system thoroughly. First of all, this dissertation models the energy-efficient routing under the unique pattern of the cross-MMN behavior. A new modeling approach, pseudo-dynamic modeling approach, is proposed to measure the energy-efficiency of the routing methods in the presence of the cross-MMN behavior. With this approach, it could be identified that the shortest-path routing and the load-balanced routing is energy-efficient in mobile networks and static networks respectively. For the MMN-ITV system with both mobile and static MMNs, an energy-efficient routing method, energy-threshold routing, is proposed to achieve the best tradeoff between them. Secondly, due to the cross-MMN behavior, neighbor discovery is executed frequently to help the new containers join the MMN, hence, consumes similar amount of energy as that of the data communication. By exploiting the unique pattern of the cross-MMN behavior, this dissertation proposes energy-efficient neighbor discovery wakeup schedules to save up to 60% of the energy for neighbor discovery. Vehicular Ad Hoc Networks (VANETs)-based inter-vehicle communications is by now growingly believed to enhance traffic safety and transportation management with low cost. The end-to-end delay is critical for the time-sensitive safety applications in VANETs, and can be a decisive performance metric for VANETs. This dissertation presents a complete analytical model to evaluate the end-to-end delay against the transmission range and the packet arrival rate. This model illustrates a significant end-to-end delay increase from non-saturated networks to saturated networks. It hence suggests that the distributed power control and admission control protocols for VANETs should aim at improving the real-time capacity (the maximum packet generation rate without causing saturation), instead of the delay itself. Based on the above model, it could be determined that adopting uniform transmission range for every vehicle may hinder the delay performance improvement, since it does not allow the coexistence of the short path length and the low interference. Clusters are proposed to configure non-uniform transmission range for the vehicles. Analysis and simulation confirm that such configuration can enhance the real-time capacity. In addition, it provides an improved trade off between the end-to-end delay and the network capacity. A distributed clustering protocol with minimum message overhead is proposed, which achieves low convergence time.