We propose repair pipelining , a technique that speeds up the repair performance in general erasure-coded storage. By carefully scheduling the repair of failed data in small-size units across storage nodes in a pipelined manner, repair pipelining reduces the single-block repair time to approximately the same as the normal read time for a single block in homogeneous environments. We further design different extensions of repair pipelining algorithms for heterogeneous environments and multi-block repair operations. We implement a repair pipelining prototype, called ECPipe , and integrate it as a middleware system into two versions of Hadoop Distributed File System (HDFS) (namely, HDFS-RAID and HDFS-3) as well as Quantcast File System. Experiments on a local testbed and Amazon EC2 show that repair pipelining significantly improves the performance of degraded reads and full-node recovery over existing repair techniques.

In this paper, we advocate the notion of "BIG" cache as an innovative abstraction for effectively utilizing the distributed storage and processing capacities of all servers in a cache network. The "BIG" cache abstraction is proposed to partly address the problem of (cascade) thrashing in a hierarchical network of cache servers, where it has been known that cache resources at intermediate servers are poorly utilized, especially under classical cache replacement policies such as LRU. We lay out the advantages of "BIG" cache abstraction and make a strong case both from a theoretical standpoint as well as through simulation analysis. We also develop the dCLIMB cache algorithm to minimize the overheads of moving objects across distributed cache boundaries and present a simple yet effective heuristic for addressing the cache allotment problem in the design of "BIG" cache abstraction.

Network measurement remains a missing piece in today's software packet processing platforms. Sketches provide a promising building block for filling this void by monitoring every packet with fixed-size memory and bounded errors. However, our analysis shows that existing sketch-based measurement solutions suffer from severe performance drops under high traffic load. Although sketches are efficiently designed, applying them in network measurement inevitably incurs heavy computational overhead.

To balance performance and storage efficiency, modern clustered file systems often first store data with replication, followed by encoding the replicated data with erasure coding. We argue that the commonly used random replication does not take into account erasure coding in its design, thereby raising both performance and availability issues in the subsequent encoding operation. We propose encoding-aware replication, which carefully places the replicas so as to (i) eliminate cross-rack downloads of data blocks during the encoding operation, (ii) preserve availability without data relocation after the encoding operation, and (iii) maintain load balancing across replicas as in random replication before the encoding operation. We conduct extensive HDFS-based testbed experiments and discrete-event simulations, and demonstrate the performance gains of encoding-aware replication over random replication.

To balance performance and storage efficiency, modern clustered file systems (CFSes) often first store data with random replication (i.e., distributing replicas across randomly selected nodes), followed by encoding the replicated data with erasure coding. We argue that random replication, while being commonly used, does not take into account erasure coding and hence will raise both performance and availability issues to the subsequent encoding operation. We propose encoding-aware replication, which carefully places the replicas so as to (i) avoid cross-rack downloads of data blocks during encoding, (ii) preserve availability without data relocation after encoding, and (iii) maintain load balancing as in random replication. We implement encoding-aware replication on HDFS, and show via tested experiments that it achieves significant encoding throughput gains over random replication. We also show via discrete-event simulations that encoding-aware replication remains effective under various parameter choices in a large-scale setting. We further show that encoding-aware replication evenly distributes replicas as in random replication.

With the explosive growth of data, enterprises increasingly adopt erasure coding on storage clusters to save storage space. On the other hand, erasure coding incurs higher performance overhead, especially during recovery. This motivates us to study the feasibility of alleviating performance overhead of erasure coding, while maintaining its storage efficiency advantage. In this paper, we study the performance issue of MapReduce when it runs on erasure-coded storage. We first review our previously proposed degraded-first scheduling, which avoids network bandwidth competition among degraded map tasks in failure mode, and hence improves the MapReduce performance over the default locality-first scheduling in MapReduce. We then show that the basic degraded-first scheduling may not work effectively when there are multiple running MapReduce jobs, and hence we propose heuristics to enhance the degraded-first scheduling design. Simulations demonstrate the performance gain of our enhanced degraded-first scheduling in a multi-job scenario. Our work makes a case that a new design of MapReduce scheduling is critical when we move to erasure-coded storage.

Data availability is critical in distributed storage systems, especially when node failures are prevalent in real life. A key requirement is to minimize the amount of data transferred among nodes when recovering the lost or unavailable data of failed nodes. This paper explores recovery solutions based on regenerating codes, which have been designed to provide fault-tolerant storage and minimum bandwidth. Existing optimal regenerating codes are designed for single node failures. We build a system called CORE, which augments existing optimal regenerating codes for the recovery of a general number of failures including single and concurrent failures. We show theoretically that CORE achieves the minimum possible bandwidth for most cases. We implement a CORE prototype and evaluate it atop an HDFS cluster testbed with up to 20 storage nodes. We demonstrate that our CORE prototype conforms to our theoretical findings and achieves bandwidth savings when compared to the conventional recovery approach based on erasure codes.

We have witnessed an increasing adoption of erasure coding in modern clustered storage systems to reduce the storage overhead of traditional 3-way replication. However, it remains an open issue of how to customize the data analytics paradigm for erasure-coded storage, especially when the storage system operates in failure mode. We propose degraded-first scheduling, a new MapReduce scheduling scheme that improves MapReduce performance in erasure-coded clustered storage systems in failure mode. Its main idea is to launch degraded tasks earlier so as to leverage the unused network resources. We conduct mathematical analysis and discrete event simulation to show the performance gain of degraded-first scheduling over Hadoop's default locality-first scheduling. We further implement degraded-first scheduling on Hadoop and conduct test bed experiments in a 13-node cluster. We show that degraded-first scheduling reduces the MapReduce runtime of locality-first scheduling.

In modern parallel storage systems (e.g., cloud storage and data centers), it is important to provide data availability guarantees against disk (or storage node) failures via redundancy coding schemes. One coding scheme is X-code, which is double-fault tolerant while achieving the optimal update complexity. When a disk/node fails, recovery must be carried out to reduce the possibility of data unavailability. We propose an X-code-based optimal recovery scheme called minimum-disk-read-recovery (MDRR), which minimizes the number of disk reads for single-disk failure recovery. We make several contributions. First, we show that MDRR provides optimal single-disk failure recovery and reduces about 25 percent of disk reads compared to the conventional recovery approach. Second, we prove that any optimal recovery scheme for X-code cannot balance disk reads among different disks within a single stripe in general cases. Third, we propose an efficient logical encoding scheme that issues balanced disk read in a group of stripes for any recovery algorithm (including the MDRR scheme). Finally, we implement our proposed recovery schemes and conduct extensive testbed experiments in a networked storage system prototype. Experiments indicate that MDRR reduces around 20 percent of recovery time of the conventional approach, showing that our theoretical findings are applicable in practice.

Data availability is critical in distributed storage systems, especially when node failures are prevalent in real life. A key requirement is to minimize the amount of data transferred among nodes when recovering the lost or unavailable data of failed nodes. This paper explores recovery solutions based on regenerating codes, which are shown to provide fault-tolerant storage and minimum recovery bandwidth. Existing optimal regenerating codes are designed for single node failures. We build a system called CORE, which augments existing optimal regenerating codes to support a general number of failures including single and concurrent failures. We theoretically show that CORE achieves the minimum possible recovery bandwidth for most cases. We implement CORE and evaluate our prototype atop a Hadoop HDFS cluster testbed with up to 20 storage nodes. We demonstrate that our CORE prototype conforms to our theoretical findings and achieves recovery bandwidth saving when compared to the conventional recovery approach based on erasure codes.

The current parallel storage systems use thousands of inexpensive disks to meet the storage requirement of applications. Data redundancy and/or coding are used to enhance data availability, for instance, Row-diagonal parity (RDP) and EVENODD codes, which are widely used in RAID-6 storage systems, provide data availability with up to two disk failures . To reduce the probability of data unavailability, whenever a single disk fails, disk recovery will be carried out. We find that the conventional recovery schemes of RDP and EVENODD codes for a single failed disk only use one parity disk. However, there are two parity disks in the system, and both can be used for single disk failure recovery. In this article, we propose a hybrid recovery approach that uses both parities for single disk failure recovery, and we design efficient recovery schemes for RDP code (RDOR-RDP) and EVENODD code (RDOR-EVENODD). Our recovery scheme has the following attractive properties: (1) “ read optimality ” in the sense that our scheme issues the smallest number of disk reads to recover a single failed disk and it reduces approximately 1/4 of disk reads compared with conventional schemes; (2) “ load balancing property ” in that all surviving disks will be subjected to the same (or almost the same) amount of additional workload in rebuilding the failed disk. We carry out performance evaluation to quantify the merits of RDOR-RDP and RDOR-EVENODD on some widely used disks with DiskSim. The offline experimental results show that RDOR-RDP and RDOR-EVENODD outperform the conventional recovery schemes of RDP and EVENODD codes in terms of total recovery time and recovery workload on individual surviving disk. However, the improvements are less than the theoretical value (approximately 25%), as RDOR-RDP and RDOR-EVENODD change the disk access pattern from purely sequential to a more random one compared with their conventional schemes.