A new RAID-x (redundant array of inexpensive disks at level x) architecture is presented for distributed I/O processing on a serverless cluster of computers. The RAID-x architecture is based on a new concept of orthogonal striping and mirroring (OSM) across all distributed disks in the cluster. The primary advantages of this OSM approach lie in: (1) a significant improvement in parallel I/O bandwidth, (2) hiding disk mirroring overhead in the background, and (3) greatly enhanced scalability and reliability in cluster computing applications. All claimed advantages are substantiated with benchmark performance results on the Trojans cluster built at USC in 1999. Throughout the paper, we discuss the issues of scalable I/O performance, enhanced system reliability, and striped checkpointing on distributed RAID-x in a serverless cluster environment. 1.

This paper presents the integration of two collective I/O techniques into the Clusterfile parallel file system : diskdirected I/O and two-phase I/O. We show that global cooperative cache management improves the collective I/O performance. The solution focuses on integrating disk parallelism with other types of parallelism: memory (by bu#ering and caching on several nodes), network (by parallel I/O scheduling strategies) and processors (by redistributing the I/O related computation over several nodes). The performance results show considerable throughput increases over ROMIO's extended two-phase I/O.

In this paper, we propose a novel Single I/O Space architecture for achieving a Single System Image (SSI) at the I/O subsystem level. This is very much desired in a scalable cluster computing environment using commodity components. Our design achieves a single address space for all blocks of data in the cluster, which can tolerate all single disk failures. While traditional approaches focused on at user-level or at distributed file subsystem level, we separate the I/O subsystem of a cluster into the file system and a set of distributed Virtual Device Drivers (VDDs). All SSI services are provided by the VDDs with unmodified file systems. Compared to previous approaches, our approach has higher transparency, better performance, lower implementation cost, higher availability, and application compatibility for I/O intensive cluster computing. 1.

This paper presents Cluster-Aware Remote Disks (CARDs), a Single System I/O architecture for cluster computing. CARDs virtualize accesses to remote cluster disks over a System Area Network. Their operation is driven by cooperative caching policies that implement a joint management of the cluster caches. All the CARDs of a given disk employ a common policy, independently of other CARD sets. CARD drivers have been implemented as Linux kernel modules which can flexibly accommodate various cooperative caching algorithms. We designed and implemented a decentralized policy called Home-based Serverless Cooperative Caching (HSCC). HSCC showed cache hit ratios over 50 % for workloads that go beyond the limit of the global cache. The best speedup of a CARD over a remote disk interface was 1.54. 1

Parallel Computing:

The I/O patterns of large scale scientific applications can often be characterized as small, non-contiguous, and regular. From a performance and power perspective, this is perhaps the worse kind of I/O for a disk. Two approaches to mitigating the mechanical limitations of disks are write-back caches and software-directed power management. Previous distributed caches are plagued by synchronization and scalability issues. The Direct Access Cache: DAChe system is a user-level distributed cached that addresses both these problems. Past work on managing disk power during run time were effective, one should be able to improve on those results by adopting a proactive scheme. 1.

Distribution of large data objects among several storage servers is a common technique to speed up access rates. In combination with parity schemes, failures of single server nodes can be tolerated, so that such systems reach a certain degree of fault tolerance. In this paper such a distributed server system is analyzed. Data objects are stored in a data layout according to RAID level 3 among disk subsystems of different computers. An access control provides concurrent up- and down-streaming of data objects to/from the distributed storage system with ensured data consistency. This consistency control is described in combination with the handling of faulty server nodes and faulty clients. Furthermore, performance is measured with several access patterns. An application of that technique is for instance a distributed video server, allowing permanently updates without interrupting access.

In this paper we present a multiple phase I/O collective operation for generic block cyclic distributions. The communication pattern is automatically generated by an inspector phase and the communication and file access phase are performed by an executor phase. The inspector phase can be amortized over several accesses. We show that our method outperforms other techniques used for parallel I/O optimizations for small access granularities. 1

Striped file systems such as the Parallel Virtual File System (PVFS) deliver high-bandwidth I/O to applications running on clusters. An open problem in the design of striped file systems is how to reduce their vulnerability to disk failures with the minimum performance penalty. In this paper we describe a novel data redundancy scheme designed specifically to address the performance issue. We demonstrate the new scheme within CSAR, a proof-of-concept implementation based on PVFS. By dynamically switching between RAID1 and RAID5 redundancy based on write size, CSAR consistently achieves the best of two worlds- RAID1 performance on small writes, and RAID5 efficiency on large writes. Using the popular parallel I/O benchmark BTIO, our scheme achieves 82 % of the write bandwidth of the unmodified PVFS. We describe the issues in implementing our new scheme in a popular striped file system such as PVFS on a Linux cluster. 1.

Abstract. This paper evaluates the impact of the parallel I/O scheduling strategy on the performance of the file access in a parallel file system for clusters of commodity computers (Clusterfile). We argue that the parallel I/O scheduling strategy should be seen as a complement to other file access optimizations like striping over several I/O servers, non-contiguous I/O and collective I/O. Our study is based on three simple decentralized parallel I/O heuristics implemented inside Clusterfile. The measurements in a real environment show that the performance of parallel file access may vary with as much as 86 % for writing and 804% for reading with the employed heuristic and with the schedule block granularity. 1