Edge computing has been developed to utilize multiple tiers of resources for privacy, cost and Quality of Service (QoS) reasons. Edge workloads have the characteristics of data-driven and latency-sensitive. Because of this, edge systems have developed to be both heterogeneous and distributed. The unique characteristics of edge workloads and edge systems have motivated EdgeBench, a workflow-based benchmark aims to provide the ability to explore the full design space of edge workloads and edge systems. EdgeBench is both customizable and representative. It allows users to customize the workflow logic of edge workloads, the data storage backends, and the distribution of the individual workflow stages to different computing tiers. To illustrate the usability of EdgeBench, we also implements two representative edge workflows, a video analytics workflow and an IoT hub workflow that represents two distinct but common edge workloads. Both workflows are evaluated using the workflow-level and function-level metrics reported by EdgeBench to illustrate both the performance bottlenecks of the edge systems and the edge workloads.

To bridge the giant semantic gap between applications and modern storage systems, passing a piece of tiny and useful information, called I/O access hints, from upper layers to the storage layer may greatly improve application performance and ease data management in storage systems. This is especially true for heterogeneous storage systems that consist of multiple types of storage devices. Since ingesting external access hints will likely involve laborious modifications of legacy I/O stacks, it is very hard to evaluate the effect and take advantages of access hints. In this article, we design a generic and flexible framework, called HintStor, to quickly play with a set of I/O access hints and evaluate their impacts on heterogeneous storage systems. HintStor provides a new application/user-level interface, a file system plugin, and performs data management with a generic block storage data manager. We demonstrate the flexibility of HintStor by evaluating four types of access hints: file system data classification, stream ID, cloud prefetch, and I/O task scheduling on a Linux platform. The results show that HintStor can execute and evaluate various I/O access hints under different scenarios with minor modifications to the kernel and applications.

Despite the existence of file systems tailored for flash and over a decade of research into flash file systems, this paper shows that no single Linux file system performs consistently well on a commodity SSD across different workloads. We define a compleat file system as one where no workloads realize less than 30% of the best file system's performance, and most, if not all, workloads realize at least 85% of the best file system's performance, across a diverse set of microbenchmarks and applications. No file system is compleat on commodity SSDs. This paper demonstrates that one can construct a single compleat file system for commodity SSDs by introducing a set of optimizations over BetrFS. BetrFS is a compleat file system on HDDs, matching the fastest Linux file systems in its worst cases, and, in its best cases, improving performance by up to two orders of magnitude. Our optimized BetrFS (i.e., v0.6) is not only compleat, it is either the fastest or within 15% of the fastest general-purpose Linux file system on a range of microbenchmarks. At best, these optimizations improve random write throughput by 6× compared to the fastest SSD file system. At worst, our file system is competitive with the other baseline file systems. These improvements translate to application-level gains; for instance, compared to other commodity file systems, the Dovecot mailserver and an rsync of the Linux source on BetrFS show speedups of 1.13 × and 1.8 ×, respectively.

Modern solid-state drives (SSDs) are integrating more internal resources to achieve higher capacity. Parallelizing accesses across internal resources can potentially enhance the performance of SSDs. However, exploiting parallelism inside SSDs is challenging owing to real-time access conflicts. In this paper, we propose a highly parallelizable I/O scheduler (PIOS) to improve internal resource utilization in SSDs from the perspective of I/O scheduling. Specifically, we first pinpoint the conflicting flash requests with precision during the address translation in the Flash Translation Layer (FTL). Then, we introduce conflict eliminated requests (CERs) to reorganize the I/O requests in the device-level queue by dispatching conflicting flash requests to different CERs. Owing to the significant performance discrepancy between flash read and write operations, PIOS employs differentiated scheduling schemes for read and write CER queues to always allocate internal resources to the conflicting CERs that are more valuable. The small dominant size prioritized scheduling policy for the write queue significantly decreases the average write latency. The high parallelism density prioritized scheduling policy for the read queue better utilizes resources by exploiting internal parallelism aggressively. Our evaluation results show that the parallelizable I/O scheduler (PIOS) can accomplish better SSD performance than existing I/O schedulers implemented in both SSD devices and operating systems.

With the rapid increase in the amount of data produced and the development of new types of storage devices, storage tiering continues to be a popular way to achieve a good tradeoff between performance and cost-effectiveness. In a basic two-tier storage system, a storage tier with higher performance and typically higher cost (the fast tier) is used to store frequently-accessed (active) data while a large amount of less-active data are stored in the lower-performance and low-cost tier (the slow tier). Data are migrated between these two tiers according to their activity. In this article, we propose a Tier-aware Data Deduplication-based File System, called TDDFS, which can operate efficiently on top of a two-tier storage environment.Specifically, to achieve better performance, nearly all file operations are performed in the fast tier. To achieve higher cost-effectiveness, files are migrated from the fast tier to the slow tier if they are no longer active, and this migration is done with data deduplication. The distinctiveness of our design is that it maintains the non-redundant (unique) chunks produced by data deduplication in both tiers if possible. When a file is reloaded (called a reloaded file) from the slow tier to the fast tier, if some data chunks of the file already exist in the fast tier, then the data migration of these chunks from the slow tier can be avoided. Our evaluation shows that TDDFS achieves close to the best overall performance among various file-tiering designs for two-tier storage systems.

Host-aware shingled magnetic recording (HA-SMR) drives deliver higher disk capacity and expose internal zones to host, but their performance can be severely degraded when subjected to nonsequential writes (NSWs). Using solid state drives (SSDs) as a cache layer to proactively cache NSWs can alleviate this performance degradation. However, existing SSD caching designs are not capable of identifying and selectively caching NSWs, which incurs ineffectual SSD write traffic and excessively wears the limited write endurance of SSD. In this paper, we propose an endurable SMR-oriented SSD Caching (SMRC) framework. SMRC leverages the intrinsic host-aware property of HA-SMR drives to filter both sequential writes and innocuous NSWs out of SSD. Besides, SMRC reorganizes the cached NSWs to evict them with cleaning-friendly write traffic to HA-SMR drives. Our experimental results show that SMRC can effectively reduce SSD write traffic and improve system performance compared with existing SSD caching designs.

In multi-tier storage systems, moving data from one tier to the next can be inefficient. And because each type of storage device has its own idiosyncrasies with respect to the workloads that it can best support, unnecessary data movement might result. In this paper, we explore a fully connected storage architecture in which data can move from any storage pool to another. We propose a Chunk-level storage-aware workload Analyzer framework, abbreviated as ChewAnalyzer, to facilitate efficient data placement. Access patterns are characterized in a flexible way by a collection of I/O accesses to a data chunk. ChewAnalyzer employs a Hierarchical Classifier [1] to analyze the chunk patterns step by step. In each classification step, the Chunk Placement Recommender suggests new data placement policies according to the device properties. Based on the analysis of access pattern changes, the Storage Manager can adequately distribute or migrate the data chunks across different storage pools. Our experimental results show that ChewAnalyzer improves the initial data placement and that it migrates data into the proper pools directly and efficiently.

Shingled Magnetic Recording (SMR) drives can benefit large-scale storage systems by reducing the Total Cost of Ownership (TCO) of dealing with explosive data growth. Among all existing SMR models, Host Aware SMR (HA-SMR) looks the most promising for its backward compatibility with legacy I/O stacks and its ability to use new SMR-specific APIs to support host I/O stack optimization. Building storage systems using HA-SMR drives calls for a deep understanding of the drive’s performance characteristics. To accomplish this, we conduct in-depth performance evaluations on HA-SMR drives with a special emphasis on the performance implications of the SMR-specific APIs and how these drives can be deployed in large storage systems. We discover both favorable and adverse effects of using HA-SMR drives under various workloads. We also investigate the drive’s performance under legacy production environments using real-world enterprise traces. Finally, we propose a novel host-controlled buffer that can help to reduce the severity of the decline in HA-SMR performance under our discovered unfavorable I/O access patterns. Without a detailed comprehensive design, we show the potential of the host-controlled buffer by a case study.

Shingled magnetic recording (SMR) increases the capacity of magnetic hard drives, but it requires that each zone of a disk be written sequentially and erased in bulk. This makes SMR a good fit for workloads dominated by large data objects with limited churn. To explore this possibility, we have developed SMORE, an object storage system designed to reliably and efficiently store large, seldom-changing data objects on an array of host-managed or host-aware SMR disks. SMORE uses a log-structured approach to accommodate the constraint that all writes to an SMR drive must be sequential within large shingled zones. It stripes data across zones on separate disks, using erasure coding to protect against drive failure. A separate garbage collection thread reclaims space by migrating live data out of the emptiest zones so that they can be trimmed and reused. An index stored on flash and backed up to the SMR drives maps object identifiers to on-disk locations. SMORE interleaves log records with object data within SMR zones to enable index recovery after a system crash (or failure of the flash device) without any additional logging mechanism. SMORE achieves full disk bandwidth when ingesting data---with a variety of object sizes---and when reading large objects. Read performance declines for smaller object sizes where inter- object seek time dominates. With a worst-case pattern of random deletions, SMORE has a write amplification (not counting RAID parity) of less than 2.0 at 80% occupancy. By taking an index snapshot every two hours, SMORE recovers from crashes in less than a minute. More frequent snapshots allow faster recovery.

For many applications of multimedia medical devices in clinical and medical issues, cloud computing becomes a very useful way. However, high energy consumption of cloud computing networks for these applications brings forth a large challenge. This paper studies the energy-efficient problem with QoS constraints in large-scale cloud computing networks. We use the sleeping and rate scaling mechanism to propose a link energy consumption model to characterize the network energy consumption. If there is no traffic on a link, we will let it be sleeping. Otherwise, it is activated and we divide its energy consumption into base energy consumption and traffic energy consumption. The former describes the constant energy consumption that exists when the link runs, while the later, which is a quadratic function with respect to the traffic, indicates the relations between link energy consumption and the traffic on the link. Then considering the relation among network energy consumption, number of active links, and QoS constraints, we build the multi-constrained energy efficient model to overcome the high energy consumption in large-scale cloud computing networks. Finally, we exploit the NSF and GEANT network topology to validate our model. Simulation results show that our approach can significantly improve energy efficiency of cloud computing networks.

Virtualization in clouds is promoting the current trend of sharing the storage with multiple tenants. This brings us two fundamental design issues when considering SSDs as a shared storage cache. (i) How can we choose the hierarchy cache model to reduce I/O latency? (ii) How can we design the dynamic cache space allocation strategy to maximize utilization of SSDs space? This paper mainly proposes the corresponding solutions to address the above two issues. (i) We design a cache-awareness model to avoid the useless network latency for querying. (ii) With using the weighted max-min fair share algorithm, measuring the weighted value of each tenant through recording the states of four multi-dimensional factors, can be beneficial to make wise decisions on SSD space allocation. Our experimental results validate that the cache-awareness model outperforms the other two models (cache-unawareness model and without cache) by 1× to 4× in latency. Meanwhile, compared to the static weighted value initializations of the max-min algorithm, our method with dynamically measuring weight value on a tenant basis can achieve much better space utilization.

Network Redundancy Elimination (NRE) aims to improve network performance by identifying and removing repeated transmission of duplicate content from remote servers. Using a Content-Defined Chunking (CDC) policy, an inline NRE process can obtain a higher Redundancy Elimination (RE) ratio but may suffer from a considerably higher computational requirement than fixed-size chunking. Additionally, the existing work on NRE is either based on IP packet level redundancy elimination or rigidly adopting a CDC policy with a static empirically-decided expected chunk size. These approaches make it difficult for conventional NRE MiddleBoxes to achieve both high network throughput to match the increasing line speeds and a high RE ratio at the same time. In this paper we present a design and implementation of an inline NRE appliance which incorporates an improved FPGA-based scheme to speed up CDC processing to match the ever increasing network line speeds while simultaneously obtaining a high RE ratio. The overhead of Rabin fingerprinting, which is a key component of CDC, is greatly reduced through the use of a record table and registers in the FPGA. To efficiently utilize the hardware resources, the whole NRE process is handled by a Virtualized NRE (VNRE) controller. The uniqueness of this VNRE that we developed lies in its ability to exploit the redundancy patterns of different TCP flows and customize the chunking process to achieve a higher RE ratio. VNRE will first decide if the chunking policy should be either fixed-size chunking or CDC. Then VNRE decides the expected chunk size for the corresponding chunking policy based on the TCP flow patterns. Implemented in a partially reconfigurable FPGA card, our trace driven evaluation demonstrates that the chunking throughput for CDC in one FPGA processing unit outperforms chunking running in a virtual CPU by nearly 3X. Moreover, through the differentiation of chunking policies for each flow, the overall throughput of the VNRE appliance outperforms one with static NRE configurations by 6X to 57X while still guaranteeing a high RE ratio.

With the development of multimedia and Internet of things technologies, technology enhanced learning applications such as smart class and smart learning at home have received more attentions from industrial and academic communities. However, pervasive wireless communications, which builds new collaborative and personalized learning patterns, have an important impact on learning and teaching. For current pervasive wireless networks, spectrum resources become increasing lack due to the introduction of a large number of newwireless technologies and thus the access of many newdevices to wireless networks for all kinds of applications such as online smart learning. How to exploit effectively spectrum resources in current pervasive wireless networks is a larger challenge. To this end, we study dynamic access problem for spectrum resources. In this paper, we analyze more complex multiple channel model with multiple primary and secondary users. We exploit the lognormal distribution to characterize primary user behaviors and use opportunistic spectrum access to obtain the sensing status of channels. To raise the channel utility, we control transmission radiuses of secondary users to make primary and secondary users simultaneously utilize the channel. Finally, we propose the dynamic access algorithm to multiple channels. Simulation results show that our approach is feasible and promising.

Shingled Magnetic Recording (SMR) technology increases the areal density of hard disk drives. Among the three types of SMR drives on the market today, Host Aware SMR (HA-SMR) drives look the most promising. In this paper, we carry out evaluation to understand the performance of HA-SMR drives with the objective of building large-scale storage systems using this type of drive. We focus on evaluating the special features of HA-SMR drives, such as the open zone issue and media cache cleaning efficiency. Based on our observations we propose a novel host-controlled indirection buffer to enhance the drive's I/O performance. Finally, we present a case study of the open zone issue to show the potential of this host-controlled indirection buffer for HA-SMR drives.

This paper proposes a deMOtion-based, fLAsh-awaRe hybrid storage cache model, named MOLAR, to effectively integrate Flash-based Solid State Disks (SSDs) into traditional DRAM-based memory storage systems. In MOLAR, a flash-aware I/O path structure is designed to adapt the asymmetric read and write performance of SSD and moreover to reduce useless write operations. A new control metric, demotion count, is proposed to wisely select the evicted data blocks from DRAM to reside in SSD. Besides, for SSD can improve internal data placement from data access hints, the Logical Block Addresses (LBAs) in SSD are grouped into the long-lived region and the short-lived region self-adaptively via a heuristic control algorithm based on the change of data block demotion count. Through trace-driven simulations, the overall hit ratio in MOLAR outperforms two traditional policies from 1.44% to 5.34%. The average write latency in SSD is reduced by 3.5 X. Moreover, write amplification is effectively reduced by about 36% in two typical flash address mapping policies.

Compared with a hash table, a Bloom Filter (BF) is more space-efficient for supporting fast matching though resulting in a controllable and acceptable false positive probability. The space size of the basic BF is predetermined based on the expected number of elements to be stored. However, we cannot predict the scale of a BF space for dynamic sets. The two existing solutions for supporting dynamic sets, Scalable BF (SBF) and Dynamic BF (DBF), still face some challenges on system performance and memory overhead.This paper presents a new BF for dynamic data sets, called Partitioned BF (Par-BF). Compared with DBF and SBF, the size and the range of the false positive probability can be calculated by a group of formulas to leverage a sweet spot between high-performance and low-overhead. Moreover, Par-BF supports parallel fast matching which can improve the overall throughput. From our trace-driven experimental results, the IOPS of Par-BF outperforms that of DBF and SBF from 6X to 10X, and from 2X to 4X, respectively. Meanwhile, through our proposed garbage collection policy, the memory overhead of Par-BF is less than half of the memory usage of SBF.

The resources of dedicated accelerators (e.g. FPGA) are still required to bridge the gap between software-based Middleboxs(MBs) and the commodity hardware. To consolidate various hardware resources in an elastic, programmable and reconfigurable manner, we design and build a flexible and consolidated framework, OpenANFV, to support virtualized accelerators for MBs in the cloud environment. OpenANFV is seamlessly and efficiently put into Openstack to provide high performance on top of commodity hardware to cope with various virtual function requirements. OpenANFV works as an independent component to manage and virtualize the acceleration resources (e.g. cinder manages block storage resources and nova manages computing resources). Specially, OpenANFV mainly has the following three features. (1)Automated Management. Provisioning for multiple Virtualized Network Functions (VNFs) is automated to meet the dynamic requirements of NFV environment. Such automation alleviates the time pressure of the complicated provisioning and configuration as well as reduces the probability of manually induced configuration errors. (2) Elasticity. VNFs are created, migrated, and destroyed on demand in real time. The reconfigurable hardware resources in pool can rapidly and flexibly offload the corresponding services to the accelerator platform in the dynamic NFV environment. (3) Coordinating with Openstack. The design and implementation of the OpenANFV APIs coordinate with the mechanisms in Openstack to support required virtualized MBs for multiple tenants.

The resources of dedicated accelerators (e.g. FPGA) are still required to bridge the gap between software-based Middleboxs(MBs) and the commodity hardware. To consolidate various hardware resources in an elastic, programmable and reconfigurable manner, we design and build a flexible and consolidated framework, OpenANFV, to support virtualized accelerators for MBs in the cloud environment. OpenANFV is seamlessly and efficiently put into Openstack to provide high performance on top of commodity hardware to cope with various virtual function requirements. OpenANFV works as an independent component to manage and virtualize the acceleration resources (e.g. cinder manages block storage resources and nova manages computing resources). Specially, OpenANFV mainly has the following three features. (1)Automated Management. Provisioning for multiple Virtualized Network Functions (VNFs) is automated to meet the dynamic requirements of NFV environment. Such automation alleviates the time pressure of the complicated provisioning and configuration as well as reduces the probability of manually induced configuration errors. (2) Elasticity. VNFs are created, migrated, and destroyed on demand in real time. The reconfigurable hardware resources in pool can rapidly and flexibly offload the corresponding services to the accelerator platform in the dynamic NFV environment. (3) Coordinating with Openstack. The design and implementation of the OpenANFV APIs coordinate with the mechanisms in Openstack to support required virtualized MBs for multiple tenants.