ActiveSort: Efficient external sorting using active SSDs in the MapReduce framework

Nowadays solid state drives (SSDs) are gaining popularity and are replacing magnetic hard disk drives (HDDs) in enterprise storage systems. As a result, extracting the maximum performance from SSDs is becoming crucial to deal with the increasing storage volume and performance needs. Active disks were introduced as a way to offload data-processing tasks from the host into disks freeing system resources and achieving better performance. In this work, we present an active SSD architecture called Self-Sorting SSD that targets to offload sorting operations which are commonly used in data-intensive and database environments and that require heavy data transfer. Processing sorting operations directly on the SSD reduces data transfer from/to the storage devices, increasing system performance and the lifetime of SSDs. Experiments on a real SSD platform reveal that our proposed architecture outperforms traditional external merge sort by up to 60.75%, reduces energy consumption by up to 58.86%, and eliminates all the data transfer overhead to compute sorted results.

Flash memory solid state drives (SSDs) have increasingly been advocated and adopted as a means of speeding up and scaling up data-driven applications. SSDs are becoming more widely available as an option in the cloud. However, when an application considers SSDs in the cloud, the best option for the application may not be immediate, among a number of choices for placing SSDs in the layers of the cloud. Although there have been many studies on SSDs, they often concern a specific setting, and how different SSD options in the cloud compare with each other is less well understood. In this paper, we describe how Google Fusion Tables (GFT) used SSDs and what optimizations were implemented to scale up its in-memory processing, clearly showing opportunities and limitations of SSDs in the cloud with quantitative analyses. We first discuss various SSD placement strategies and compare them with low-level measurements, and propose SSD-placement guidelines for a variety of cloud data services. We then present internals of our column engine and optimizations to better use the performance characteristics of SSDs. We empirically demonstrate that the optimizations enable us to scale our application to much larger datasets while retaining the low-latency and simple query processing architecture.

Solid State Drives (SSDs) have become industry standard for data storage due to their compact form factor, high data transfer speeds, and low power consumption. However, each SSD manufacturer uses their own proprietary firmware, meaning the user does not know how data is stored on the SSD. The primary contribution of this work is the demonstration of the capability to classify activity (read and write) on an SSD versus non-activity (idle) using the temperature side-channel. In addition, we present promising initial results using temperature signatures obtained from an infrared camera to classify read, write, and idle operations. Read and write operations were evaluated at 100 MB, 500 MB, and 1000 MB on four industry standard M.2 SSDs. This study demonstrated the ability to determine when an SSD was inactive with 100% recall, while being able to identify an active SSD with at least 94 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> sensitivity for each of the four SSDs. We were able to classify operations with an accuracy of 80% on three of the four SSDs.

When originally introduced, flash based solid state drives (SSD) exhibited a very high random read throughput with low sub-millisecond latencies. However, in addition to their steep prices, SSDs suffered from slow write rates and reliability concerns related to cell wear. For these reasons, they were relegated to a niche status in the consumer and personal computer market. Since then, several architectural enhancements have been introduced that led to a substantial increase in random write operations as well as a reasonable improvement in reliability. From a purely performance point of view, these high I/O rates and improved reliability make the SSDs an ideal choice for enterprise On-Line Transaction Processing (OLTP) applications. However, from a price/performance point of view, the case for SSDs may not be clear. Enterprise class SSD Price/GB, continues to be at least 10x higher than conventional magnetic hard disk drives (HDD) despite considerable drop in Flash chip prices. We show that a complete replacement of traditional HDDs with SSDs is not cost effective. Further, we demonstrate that the most cost efficient use of SSDs for OLTP workloads is as an intermediate persistent cache that sits between conventional HDDs and memory, thus forming a three-level memory hierarchy. We also discuss two implementations of such cache: hardware or software. For the software approach, we discuss our implementation of such a cache in an inhouse database system. We also describe off-the shelf hardware solutions. We will develop a Total Cost of Ownership (TCO) model for All-SSD and All-HDD configurations. We will also come up with a modified OLTP benchmark that can scale IO density to validate this model. We will also show how such SSD cache implementations could increase the performance of OLTP applications while reducing the overall system cost.

Rowhammer has been shown to be an extensive attack vector. In the years since its discovery, numerous exploits have been shown, attacking a wide range of targets from kernels, through web browsers to machine learning models. These attacks were not always mounted from code running on the CPU of a system. Various devices peripheral to the CPU, like GPUs or networks cards can cause Rowhammer bit flips through DMA accesses to the main memory. In this work, we take a look at solid state drives (SSDs) and if they can be exploited as confused deputies to perform Rowhammer attacks. With the introduction of NVMe, a standardized protocol that allows SSDs to communicate directly over PCIe with the CPU, SSDs have reached performance numbers of a million input/output operations per second. PCIe also enables SSDs to use DMA for direct accesses to the main memory. This lead to the introduction of the host memory buffer (HMB) feature, that allows SSDs to use a small fraction of the host DRAM. We are the first that reverse engineer how different SSDs utilize this host memory buffer and answer the question if the accesses from the SSD to the HMB are a potential attack vector to cause Rowhammer bit flips. Our analysis of three SSDs shows, that bit flips in the HMB cause the SSDs to lock up, which results in a denial of service or, even worse, data loss. We also show how we can cause frequent accesses from the SSD to the HMB on all three SSDs. On one SSD, we reach 5000 DRAM accesses per refresh interval. We measure the Rowhammer impact of these accesses and show that they are effectively hammering the DRAM. However, 5000 DRAM accesses are not enough to cause Rowhammer bit flips, even on modern, highly vulnerable DRAM.

Recently, flash-based solid state drives (SSDs) have become a primary storage solution due to their advantages over hard-disk drives. Nonetheless, SSD management presents unique challenges. First, SSDs update data by writing a new copy to a clean slot, rather than overwriting old data. Second, SSDs support cleaning of entire blocks, while slots cannot be cleaned individually. Third, when cleaning a block, any valid data it stores must first be rewritten to another location, which adversely affects the SSD endurance and throughput. In this work, we address the SSD management problem, where the objective is to minimize the number of rewrites in SSDs. We approach the problem from a theoretical perspective, analyzing algorithms in a worst-case fashion. Motivated by recent advances in machine learning, we consider a learning-augmented setting where the algorithm has access to a predictive oracle, with performance guarantees expressed as a function of the error in the oracle's output. Our main contribution is Gladiator, a novel online algorithm that optimally leverages predictions to enhance the performance of SSDs. Compared to prior work, Gladiator requires no additional non-volatile memory and is less sensitive to prediction errors. We further demonstrate the effectiveness of Gladiator under reasonable assumptions on the input distribution, and extend it to an extremely efficient offline algorithm. Empirical results confirm its superiority over state-of-the-art practical solutions across diverse SSD workloads.

Solid state drives (SSDs) are increasingly deployed to construct storage arrays (RAIDs) in enterprise environments. The design decisions in RAID are traditionally devised for HDD RAIDs, which cannot fully exploit the characteristics of SSDs. In particular, SSD lacks the ability to update pages in-place. Random writes in traditional parity-based RAIS (SSD RAID) systems that has static striping result in significantly more writes, degraded performance, and shortened SSD lifetime. By dynamically forming full stripes, log-based design originally proposed in HDD RAID can mitigate the write-hole problem caused by random writes. However, it needs a directory to record locations for all data blocks, resulting in large space overhead and consequently sacrificing addressing efficiency. In this paper, we propose CD-RAIS, a compromise between static striping and dynamic striping. CD-RAIS groups requests that are from different SSD drives and places their corresponding unnecessarily consecutive logical blocks in one stripe. It mitigates the write-hole problem, meanwhile remains the same addressing efficiency as static striping. To enable dynamic data striping fit SSDs, CD-RAIS performs lazy data invalidation and consolidates updates to parity blocks. CD-RAIS greatly alleviates the write request increase due to parity block update in RAIS. Our experimental results show that, for random write dominated workloads, CD-RAIS achieves 65% response time improvement and 31% longer SSD lifespan over traditional RAIS schemes.

Solid-state drive (SSD) with flash memory as the storage medium are being widely used in various data storage systems. SSD data compression means that data is compressed before it is written to Not-And (NAND) Flash. Data compression can reduce the amount of data written in NAND Flash and improve the performance and reliability of SSDs. At present, the main problem facing data compression of SSD is how to improve the efficiency of data compression and decompression. In order to improve the performance of data compression and decompression, this study proposes a method of SSD data deduplication based on zone division. First, this study divides the storage space of the SSD into zones and divides them into one hot zone and multiple cold zones according to the different erasing frequency. Second, the data in each zone is divided into hot data and cold data according to the number of erasures. At the same time, the address mapping table in the hot zone is loaded into the cache. Finally, when there is a write or read request, the SSD will selectively compress or decompress the data according to the type of different zones. Through simulation tests, the correctness and effectiveness of this study are verified. The research results show that the data compression rate of this research result can reach 70–95%. Compared with SSD without data compression, write amplification is reduced by 5 to 30%, and write latency is reduced by 5 to 25%. The research results have certain reference value for improving the performance and reliability of SSD.

Solid State Drives (SSDs) have become popular for mobile devices and PCs nowadays due to their advantages compared with conventional hard disk drives. However, they suffer from endurance problems. An effective approach to increase the endurance of an SSD is to improve the management algorithm of the RAM buffer, a cache-like structure inside the SSD. There are several previous studies of RAM buffer management algorithms which improve the I/O performance and endurance of SSDs by reordering the write operations and exploiting the temporal locality of the write stream. We observe that the sequential write patterns of the write stream can be exploited to reduce the number of expensive erase operations. Using this insight, we propose a new effective method for RAM buffer management algorithms to enhance the endurance of SSDs. We evaluate our proposed method by software simulation and the result shows that the proposed method reduces the number of erasures and average response time compared to the best existing method by up to 17.3% and 16.9% respectively.

Recently, solid state drives (SSDs) are replacing hard disk drives (HDDs) in datacenter storage systems in order to reduce power consumption and improve I/O performance. Additionally, in order to mitigate the performance bottleneck at I/O interface between host and SSD, the PCIe-leveraging NVMe SSD is emerging for datacenter SSDs. The NVMe interface supports the I/O virtualization mechanism called single root I/O virtualization (SR-IOV), which is a device self-virtualization technique for supporting direct paths from virtual machines (VMs) to I/O devices. Multiple virtual machines can share an SR-IOV-supporting physical device without intervention of virtual machine monitor. SR-IOV-supporting SSD should provide a device-level scheduler which can schedule the requests from multiple VMs considering performance isolation and fairness. In this paper, we propose a workload-aware budget compensation scheduling algorithm for the device-level request scheduler. To guarantee the performance isolation, the device-level scheduler estimates the contribution on the garbage collection (GC) cost of each virtual machine in the SSD device. Based on the estimated GC contributions, the budget of each VM is compensated for performance isolation. We experimented the effects of the proposed technique with an SSD simulator. The experiments showed that the scheduler can guarantee the performance isolation when multiple VMs share an NVMe SSD with different workloads.

As the flash-based solid-state drives(SSDs) gradually replace the mechanical hard disk drives(HDDs) as the mainstream storage, unlike the HDDs, SSDs have rich internal parallelism, which makes it have the excellent characteristics that HDDs do not have. External mergesort, as the classical algorithm of external sorting adopted in many systems and algorithms, has an important impact on the overall performance. Therefore, it is of great significance to optimize and improve the efficiency of external mergesort algorithm. The research work on optimizing raw external mergesort algorithm on SSDs is relatively few. Thus, aiming at the external mergesort problem, based on the characteristics of SSDs, this paper proposes the SortDecision algorithm which can calculate its optimal execution scheme, including merging way, read buffer size, and write buffer size which determine the execution process of external mergesort. Exploiting the above optimal execution scheme, external mergesort can obtain better efficiency. In the SortDecision algorithm, external mergesort problem on SSDs is formalized and transformed into a piecewise convex optimization problem. Then, the optimal external mergesort scheme is obtained by enumerating the solutions of each subconvex problem. The experimental results show that the external mergesort proceeds guided by SortDecision algorithm can achieve a speedup of 1 $\sim $ 6.7 compared to the traditional external mergesort algorithm in the case of limited memory provided. The richer the internal parallelism resources inside SSDs, the better the effect of SortDecision’s acceleration.

Solid-State Drives (SSDs) play a crucial role in today's storage systems. They are appended into the Hard-Disk Drives (HDDs) storage systems to improve performance. They provide high IO rate and low latency, which makes them a perfect candidate for analytic-based workloads such as MapReduce. Defining an efficient SSD deployment strategies for MapReduce workloads is a challenging task: SSDs are costly and have limited capacity, the workloads have a big process data size, and the platform has a unique workflow nature. The goal of the work is to establish performance and cost relationship between SSD approaches and MapReduce workloads. In our setup, MapReduce workloads were executed with two SSD approaches of tiering and caching each with two setups: compress and uncompress. Our results showed that by using SSD as a tier, MapReduce workload performs better by up to 66% and increased SSD lifespan by around 20% when comparing with cache approach. We also observed that applying compression on the tier approach enhanced the lifespan by 60% but reduced lifespan of cache tier by 50%.

Mining large graphs has now become an important aspect of multiple diverse applications and a number of computer systems have been proposed to provide runtime support. Recent interest in this area has led to the construction of single machine graph computation systems that use solid state drives (SSDs) to store the graph. This approach reduces the cost and simplifies the implementation of graph algorithms, making computations on large graphs available to the average user. However, SSDs are slower than main memory, and making full use of their bandwidth is crucial for executing graph algorithms in a reasonable amount of time. In this paper, we present PrefEdge, a prefetcher for graph algorithms that parallelises requests to derive maximum throughput from SSDs. PrefEdge combines a judicious distribution of graph state between main memory and SSDs with an innovative read-ahead algorithm to prefetch needed data in parallel. This is in contrast to existing approaches that depend on multi-threading the graph algorithms to saturate available bandwidth. Our experiments on graph algorithms using random access show that PrefEdge not only is capable of maximising the throughput from SSDs but is also able to almost hide the effect of I/O latency. The improvements in runtime for graph algorithms is up to 14× when compared to a single threaded baseline. When compared to multi-threaded implementations, PrefEdge performs up to 80% faster without the program complexity and the programmer effort needed for multi-threaded graph algorithms.

Redundant arrays of independent disk (RAID) is an efficient approach to relieve reliability sacrifice caused by aggressive scale-out of solid state drives (SSDs). Unfortunately, RAID is unfriendly to SSDs due to redundant parity write and data rebuilding. This paper proposes a wear aware RAID design for SSDs, called WARD , which: 1) adaptively organizes RAID stripes according to real-time interblock unbalanced wear for relieving high performance and storage overhead caused by parity data and 2) migrates blocks about to break in advance and leaves these blocks unused to reduce data rebuilding overhead. An efficient block wear detection scheme is employed to detect block wear during the whole lifetime of SSDs. Beginning with a large stripe width RAID instead of the redundant worst-case RAID, WARD reorganizes RAID stripes once wear blocks with high bit error rates come out. WARD divides the original stripe into several short width RAID stripes according to the number of wear blocks and separates all wear blocks into different stripes. This not only reduces parity redundancy but also provides high reliability to avoid more than RAID recoverable error-prone chunks remaining in one stripe. For high wear blocks tending to wear-out, data in them are migrated in advance and then the blocks are left unused, which efficiently avoids performance shock caused by data rebuilding. A reliability model considering interblock unbalanced wear is proposed and reveals that WARD provides a high and stable reliability and greatly prolongs the lifetime of SSDs. Comprehensive experiments based on an SSDsim derivative simulator are carried out and experiment results show that WARD considerably improves system performance compared to the worst-case RAID.

While an SSD (Solid State Drive) has been widely used for storage in many computing systems due to its small average latency, how to provide timing guarantees of a delay-sensitive (real-time) task on a real-time system equipped with an SSD has not been fully explored. A recent study has proposed a work-constraining I/O scheduler, called K2, which has succeeded in reducing the tail latency of a real-time task at the expense of compromising the total bandwidth for real-time and non-real-time tasks. Although the queue length bound parameter of the K2 scheduler is a key to regulate the tradeoff between a decrease in the tail latency of a real-time task and an increase in penalty of the total bandwidth, the parameter’s impact on the tradeoff has not been thoroughly investigated. In particular, no studies have addressed how the case of a fully occupied SSD that incurs garbage collection changes the performance of the K2 scheduler in terms of the tail latency of the real-time task and the total bandwidth. In this paper, we systematically analyze the performance of the K2 scheduler for different I/O operation types, based on experiments on Linux. We investigate how the performance is changed on a fully occupied SSD due to garbage collection. Utilizing the investigation, we draw general guidelines on how to select a proper setting of the queue length bound for better performance. Finally, we propose how to apply the guidelines to achieve target objectives that optimize the tail latency of the real-time task and the total bandwidth at the same time, which has not been achieved by previous studies.