The aggressive scaling of technology may have helped to meet the growing demand for higher memory capacity and density, but has also made DRAM cells more prone to errors. Such a reality triggered a lot of interest in modeling DRAM behavior for either predicting the errors in advance or for adjusting DRAM circuit parameters to achieve a better tradeoff between energy efficiency and reliability. Existing modeling efforts may have studied the impact of few operating parameters and temperature on DRAM reliability using custom FPGAs setups, however they neglected the combined effect of workloadspecific features that can be systematically investigated only on a real system.In this paper, we present the results of our study on workloaddependent DRAM error behavior within a real server considering various operating parameters, such as the refresh rate, voltage and temperature. We show that the rate of single-and multi-bit errors may vary across workloads by 8x, indicating that program inherent features can affect DRAM reliability significantly. Based on this observation, we extract 249 features, such as the memory access rate, the rate of cache misses, the memory reuse time and data entropy, from various compute-intensive, caching and analytics benchmarks. We apply several supervised learning methods to construct the DRAM error behavior model for 72 servergrade DRAM chips using the memory operating parameters and extracted program inherent features. Our results show that, with an appropriate choice of program features and supervised learning method, the rate of single-and multi-bit errors can be predicted for a specific DRAM module with an average error of less than 10.5 %, as opposed to the 2.9x estimation error obtained for a conventional workload-unaware error model. Our model enables designers to predict DRAM errors in advance for less than a second and study the impact of any workload and applied software optimizations on DRAM reliability.

The main memory in today's systems is based on DRAMs, which may offer low cost and high density storage for large amounts of data but it comes with a main drawback; DRAM cells need to be refreshed frequently for retaining the stored data. The refresh rate in modern DRAMs is set based on the worst-case retention time without considering access statistics, thereby resulting in very frequent refresh operations. Such high refresh rate leads eventually to large power and performance overheads, which are increasing with higher DRAM densities. However, such high refresh rates may not even required due to extremely low probability of the actual occurrence of the assumed worst-case scenarios, or due to the implicit refresh operation that occur during every memory access, a feature that has not been yet been studied in depth. In this paper, we enhance the state-of-the-art by systematically exploiting the implicit refresh of memory access for relaxing the refresh rate, while minimizing the resulting memory errors. This is achieved by modifying the algorithmic parameters that influence the access patterns such that all stored data are being touched within a target time interval that is necessary for meeting a target error rate. The proposed method is applied to stencil-based algorithms which represent a wide class of algorithms used in numerical analysis, image processing and cellular automata applications. The efficacy of the proposed method is demonstrated on an off-the-shelf server running a fully fledged Linux OS and results show that it is even possible to completely disable DRAM refresh with minor quality loss.

No abstract

No abstract

Improving energy efficiency of the memory subsystem becomes increasingly important for all digital systems due to the rapid growth of data. Many recent schemes have attempted to reduce the DRAM power by relaxing the refresh rate, which may negatively affect the DRAM reliability. To optimize the trade-offs between power and reliability, existing studies rely on experimental setups based on FPGAs and the use of few known data-patterns for exciting rare worst-case circuit reliability effects. However, by doing so, existing studies may be missing to capture the real DRAM behavior within a commodity server with a fully fledged OS. In this paper, we develop an experimental framework based on a state-of-the-art 64-bit ARM based server with Linux OS, in which we enabled the characterization of 72 DRAM chips under relaxed refresh period and various temperatures controlled by a unique thermal testbed. We evaluate the DRAM reliability running singleand multi-threaded HPC workloads on such a commodity server with a fully-fledged Linux OS and a typical multilevel memory hierarchy. In fact, our results show that the manifested Word-Error-Rate under relaxed refresh period varies among the workloads and can be different from the one estimated by the few known fixed data-patterns that were conventionally used in all existing studies. We also discover that the error rates incurred by the execution of the HPC workloads may vary within a program run. Finally, our study shows that the refresh period can be relaxed by 35× leading to 11.2 % power savings on average, while avoiding any system disruption, since the available error-correcting-codes were able to correct all incurred errors up to 60 • C.

No abstract

The explosive growth of Internet-connected devices will soon result in a flood of generated data, which will increase the demand for network bandwidth as well as compute power to process the generated data. Consequently, there is a need for more energy efficient servers to empower traditional centralized Cloud data-centers as well as emerging decentralized data-centers at the Edges of the Cloud. In this paper, we present our approach, which aims at developing a new class of micro-servers-the UniServer-that exceed the conservative energy and performance scaling boundaries by introducing novel mechanisms at all layers of the design stack. The main idea lies on the realization of the intrinsic hardware heterogeneity and the development of mechanisms that will automatically expose the unique varying capabilities of each hardware component within commercial microservers and allow their operation at new extended operating points. Low overhead schemes are employed to monitor and predict the hardware behavior and report it to the system software. The system software including a virtualization and resource management layer is responsible for optimizing the system operation in terms of energy or performance, while guaranteeing non-disruptive operation under the extended operating points. Our characterization results on a 64-bit ARMv8 micro-server in 28nm process reveal large voltage margins in terms of Vmin variation among the 8 cores of the CPU chip, among 3 different sigma chips, and among different benchmarks with the potential to obtain up-to 38.8% energy savings. Similarly, DRAM characterizations show that refresh rate and voltage can be relaxed by 43x and 5%, respectively, leading to 23.2% power savings on average.

In this paper, we present the implementation of a heterogeneous-reliability DRAM framework, Shimmer, on a commodity server with a fully fledged OS. Shimmer enables splitting of DRAM into multiple domains with varying reliability and allocation of data depending on their criticality. Compared to existing studies which use simulators, we consider practical restrictions stemming from the real hardware and investigate methods to overcome them. In particular, we reveal that the implementation of the heterogeneous-reliability memory framework requires disabling of the hardware memory interleaving, which results in a significant degradation of the system performance. To overcome the induced performance loss, we develop a software-based interleaving. We evaluate the performance, power and energy of the server using 35 benchmarks across three memory configurations: the baseline configuration; with disabled hardware memory interleaving and Shimmer with software-based memory interleaving. Our results show that Shimmer introduces a minor 6% performance overhead, while reducing the average DRAM power by 19.9% when memory operates under relaxed refresh rate and lowered memory supply voltage. As one of our main contributions we demonstrate that a heterogeneous-reliability framework based on Shimmer can be realized on a commodity server and save 9.1% of the total processor and memory energy.