Sharif Digital Repository

Solid-State Drives (SSDs) provide high performance, low power consumption, and hence are widely used in the I/O cache of enterprise storage systems and high performance servers. Although employing high performance SSDs in the I/O cache of storage systems significantly improves system performance, it comes with significant reliability threats for write operations upon power failures. In order to provide high level of reliability, storage architectures employ SSDs in I/O cache with RAID-1 configuration, however, such configuration only partially mitigates the reliability issues of the SSDs. In this thesis, we analyze the reliability of SSD-based I/O caches against power outage and high... 

Data storage systems are an essential part of any Information Technology (IT) infrastructure. Unavailability of these systems can directly lead to service disruption to customers and can be very costly in enterprise applications such as ebanking, online reservation systems, and mobile services. Modeling availability of disk subsystem is a major step in design of data storage systems. Previous modeling techniques have mainly focused on data unavailability or data loss due to double or multiple disk failures. In this thesis, we present both simulation and analytical modeling techniques to investigate the effect of serviceability on the overall availability of disk subsystems. The proposed... 

In recent years, emerging Non-Volatile Memories (NVMs) have become promising alternatives for existing memory technologies. Due to shortcomings of SRAM memory in nanometer era,NVMs such as Phase-Change Memory (PCM) can be used in configuration memories of Field-Programmable Gate Arrays (FPGAs). Despite prominent features of emerging NVMs, they suffer from high write-power, high write-latency, and limited number of reliable write opera-tions. In addition, a dedicated Peripheral Circuit (PC) which is required to convert the NVM state to the equivalent voltage level can impose significant area and power overheads to FPGAs.In this thesis, a reliable power-efficient hybrid architecture employing... 

With increasing rate of digital information in the world, design, configuration, and networking of enterprise storage systems has become an essential part in designing data centers. The performance of data storage systems in serving incoming requests is one of the major parameters of such systems. A major metric to measure performance is response time. This parameter is, in particular, crucial in enterprise applications such as financial, credit, multimedia, and real-time applications. A common approach to enhance the performance of enterprise storage systems is improving the hit ratio of the system global memory using prefetching technique. Using prefetching technique, a data block is... 

Nowadays, enterprise storage systems are widely used in data centers such as databases, servers, and bank institutes. Due to mechanical activity of disk drives, these systems are one of the most power consumptive components in data centers. Prefetching and RAID, which are employed to enhance performance, are widely used in data storage systems. These two methods, however, impose significant energy overhead to the system due to the increased disk activity. In this thesis, we first elaborate shortcomings of prefetching and RAID methods. Then, we propose a new energy efficient and RAID-compatible prefetching method (LPTAP), which aims at improving performance and energy consumption. To evaluate... 

Performance and reliability are the major concerns in design of data storage systems. Using Redundant Array of Independent Disks (RAID) can enhance both performance and reliability. A commonly used configuration in RAID arrays is erasure codes to further enhance system reliability. Due to promising advantages of Solid-State Drives (SSDs)such as low power, shock resistance, and high I/O performance, designers have been encouraged to replace Hard Disk Drives (HDDs) by SSDs in RAID arrays. SSDs, however, suffer from limited endurance due to erase-before-write operation.In this thesis, we investigate the effect of stripe unite size and the number of disks of a RAID array on the endurance of SSDs... 

One of the most intrinsic challenges of flash-based Solid State Drives (SSDs) is erasebefore-write limitation and the limited endurance of flash chips. Wear leveling and garbage collection are two mechanisms implemented in SSD’s controller to enhance endurance and performance. While wear leveling attempts to distribute erasures across all blocks in an even manner, it imposes a new security challenge on SSD,which leads to the presence of invalid data blocks in flash chips. Therefore, some data blocks that are logically deleted by the user are still available in flash chips and can be recovered by software or hardware recovery tools.
In this paper, a new criteria named vulnerability time... 

Signifcant downscaling of CMOS technology to provide high performance along with low power consumption has been accompanied by reliability degradations such as increased noise sensitivity and reliability concern due to aging phenomena (transient and permanent failuers). In this regard, with huge number of configuration bits, Field-Programmable Gate Arrays (FPGAs) are more susceptible to aging since aging not only degrades the performance, but also it may additionally result in corrupting the configuration cells and thus causing permanent circuit malfunctioning. While several works have investigated the aging effects in Look-Up Tables (LUTs), the routing fabric of these devices is seldom... 

Significant increase of static power with downscaling of transistor feature size and threshold voltage has lead to the end of Dennard scaling. This obstacle has put a Power Wall to further integration of CMOS technology in Field Programmable Gate Arrays (FPGAs). An efficient solution to cope with this obstacle is to apply power gating to the inactive fractions of a single die, referred to as Dark Silicon. Previous studies employing power gating on SRAM-based FPGAs have primarily focused on using large-input Look-up Tables (LUTs) which suffer from poor logic utilization, and subsequently, limiting the benefits of power gating techniques. This thesis proposes a heterogeneous Power-Efficient... 

Nowadays, embedded processors are widely used in wide range of domains from low-power to safety-critical applications. By providing prominent features such as variant peripheral support and flexibility to partial or major design modifications, Field-Programmable Gate Arrays (FPGAs) are used in industry for implementing either an entire embedded system or a Hardware Description Language (HDL)-based processor, known as soft-core processor. FPGA-based designs, however, suffer from high power consumption, large die area, and low performance that hinders common use of soft-core processors. In this thesis, we present an efficient reconfigurable architecture to implement embedded processors in... 

Emerging Solid State Drives (SSDs) have performance advantages over traditional Hard Disk Drives (HDDs). Higher price per capacity and limited lifetime, however, prevents enterprise data centers to entirely replace HDD-based storage subsystems with SSDs. Thus,SSD-based caching has been widely employed in data centers to benefit from higher performance of SSDs while minimizing overall cost. Input/Output (I/O) workloads exhibit unpredictable and highly random behavior which makes conventional algorithms such as Least Recently Used (LRU) not able to provide high hit ratio as they employ linear localities.In addition to poor performance, such algorithms also shorten SSD lifetime with unnecessary... 

With increases in amount of data being processed and generated, the need for highperformance data storage systems is essential. These systems must keep up with request throughputs as high as millions of requests per second. Solid-State Drives (SSD) lack mechanical parts and have lower read and write latency. SSDs have key differences with traditional spinning drives in having a fixed read/write granularity, erase before write and requiring wear-leveling and garbage collection operations. These operations have significant impact on drive performance and also frequency of these operations depends on the workload characteristics, such as randomness. In this thesis, a high-level estimator of SSD... 

FPGA is a suitable substrate for implementation of embedded systems, mobiles, and hand-held devices due to cost reduction for \emph{Non-Recurring Engineering (NRE)}, short time to market, design flexibility, and reprogramming capability. Significant downscaling of CMOS technology feature size has led to static power growth rate, which is a limiting factor in further scaling. Previous studies aimed at reducing power consumption, mainly have focused on the power consumption of logical resources. However, proposing a low power architecture in routing network affects the power consumption of FPGAs significantly, because of the dominant power consumption in the routing network. This thesis... 

Quality of Service (QoS) has recently become one of the most important challenges in data storage systems due to the growing use of data intensive applications in enterprise and mainstream systems. Flash-based Solid-State Drives (SSDs) are high-performance storage devices used in data storage systems to fulfill the performance demands of different applications using shared data storage system. Flash-based storage systems have limited endurance and poor write performance, resulting in quality of service issues such as unfairness. Different techniques such as I/O caching can be used to address QoS in storage systems but they adversely affect system endurance. To our knowledge, previous studies... 

Spin-Transfer Torque Magnetic RAM (STT-MRAM) is the most promising nonvolatile memory to replace SRAM technology in the Last-Level Cache (LLC) due to its benefits such as high density, near-zero cell leakage, and immunity to soft errors.However, due to its high retention failure and read disturbance rates in the downscaled technologies and the low data access rate in the LLC and the high number of read accesses, retention failure and read disturbance have become the main reliability challenges for STT-MRAM cache memory. The existing approaches to overcome these challenges impose significant area and performance overhead or adversely affect the other types of failures. In this thesis, we... 

Due to reduced Non-Recurring Engineering (NRE) costs, shorter time to market, design flexibility, and reprogramming capability of Field-Programmable Gate Arrays (FPGAs) as compared to Application-Specific Integrated Circuits (ASICs), FPGAs has been raised as a suitable substrate for implementation of digital systems. However, the high flexibility of reconfigurable devices leads to great power consumption, chip area, and reliability difference between ASICs and FPGAs. In addition, with the advent of multi-tenant FPGAs in cloud computing environments, it has been shown that crosstalk side-channel attack, which can be used by a malicious IP to leak valuable information, has become an urgent... 

In new generations of data storage systems, there is a tendency to integrate processing and storage elements. This is currently recognized with the commercial name of Hyperconverged Infrastructure, or HCI for short. The goal of this infrastructure is to meet the increasing requirements of data-intensive applications. As a new paradigm, HCI needs to incorporate different performance improvement techniques such as I/O caching. Contemporary I/O caching schemes are optimized for traditional storage architectures and do not prove optimal in a Hyperconverged Infrastructure. The main challenge of I/O cache in this kind of infrastructure is the distribution of its resources over an array of nodes... 

The lifespan of hard disks greatly affects their performance and reliability. Ensuring the freshness of disks during the procurement process for hardware manufacturing companies is crucial for maintaining efficient production systems. Predicting the approximate lifespan of hard disks involves analyzing their characteristics and operational history. Relying solely on SMART reports for estimating disk lifespan is not reliable due to the potential firmware tampering. Previous studies have primarily focused on SATA disks, investigating the relationship between failures and SMART parameter values, while neglecting SAS disks. However, in this research, we collected SMART reports from approximately... 

In the past two decades, there has been a significant performance enhancement in processors by leveraging nano-scale semiconductor technologies and micro-architectural techniques. At the same time, there has been a limited performance improvement in storage devices. This performance gap results in a performance bottleneck in computer systems. To fill this gap, Solid-State Disks (SSDs) has been proposed in the previous work. Due to not using mechanical parts, SSDs can provide higher performance and lower power consumption compared to hard disk drives. Typically, SSDs use flash memory chips to store user data. Flash memory has some shortcomings such as limited endurance and low write... 

Solid-State Drives (SSDs) are replaced to a great portion of Hard Disk Drives (HDDs) in personal computers, servers, and supercomputers due to their high reliability, shock resistance, low power consumption, and high performance. SSDs utilize either NAND or NOR flash chips in the drive backend logic to persistently store user data. Since NAND/NOR flash chips incur from limited number of write endurance and high write/erase operation latency, they use a controller called Flash Translation Layer (FTL) to alleviate these limitations. As the performance of commercially off-the-shelf (COTS) SSDs is tuned for few mainstream applications, COTS SSDs provide limited throughput for variety of...