Cloud computing centers face the key challenge of provisioning diverse virtual machine instances in an elastic and scalable manner. To address this challenge, we have performed an analysis of VM instance traces collected at six production data centers during four months. One key finding is that the number of instances created from the same VM image is relatively small at a given time and thus conventional file-based p2p sharing approaches may not be effective. Based on the understanding that different VM image files often have many common chunks of data, we propose a chunk-level Virtual machine image Distribution Network (VDN). Our distribution scheme takes advantage of the hierarchical network topology of data centers to reduce the VM instance provisioning time and also to minimize the overhead of maintaining chunk location information. Evaluation shows that VDN achieves as much as 30-80× speed up for large VM images under heavy traffic.

MapReduce offers an ease-of-use programming paradigm for processing large data sets, making it an attractive model for distributed volunteer computing systems. However, unlike on dedicated resources, where MapReduce has mostly been deployed, such volunteer computing systems have significantly higher rates of node unavailability. Furthermore, nodes are not fully controlled by the MapReduce framework. Consequently, we found the data and task replication scheme adopted by existing MapReduce implementations woefully inadequate for resources with high unavailability.

Stream processing is widely used by today's applications such as financial data analysis and disaster response. In distributed stream processing systems, machine fail-stop events are handled by either active standby or passive standby. However, existing high availability (HA) schemes have not sufficiently addressed the situation when a machine becomes temporarily unavailable due to data rate spikes, intensive analysis or job sharing, which happens frequently but lasts for short time. It is not clear how well active and passive standby fare against such transient unavailability. In this paper, we first critically examine the suitability of active and passive standby against transient unavailability in a real testbed environment. We find that both approaches have advantages and drawbacks, but neither is ideal to provide fast recovery at low overhead as required to handle transient unavailability. Based on the insights gained, we propose a novel hybrid HA method that switches between active and passive standby modes depending on the occurrence of failure events. It presents a desirable tradeoff that is different from existing HA approaches: low overhead during normal conditions and fast recovery upon transient or permanent failure events. We have implemented our hybrid method and compared it with existing HA designs with comprehensive evaluation. The results show that our hybrid method can reduce two-thirds of the recovery time compared to passive standby and 80% message overhead compared to active standby, allowing applications to enjoy uninterrupted processing without paying a high premium.

As an important technique to hide disk I/O latency, prefetching has been widely studied, and dynamic adaptive prefetching techniques have been deployed in diverse storage environments. However, two issues are not well addressed by previous research: (1) how to handle the prefetching resource allocation between concurrent sequential access streams with different request rates, and (2) how to coordinate prefetching at multiple levels in the data access path.

The multi-level storage architecture has been widely adopted in servers and data centers. However, while prefetching has been shown as a crucial technique to exploit the sequentiality in accesses common for such systems and hide the increasing relative cost of disk I/O, existing multi-level storage studies have focused mostly on cache replacement strategies. In this paper, we show that prefetching algorithms designed for single-level systems may have their limitations magnified when applied to multi-level systems. Overly conservative prefetching will not be able to effectively use the lower-level cache space, while overly aggressive prefetching will be compounded across levels and generate large amounts of wasted prefetch. We take an innovative approach to this problem: rather than designing a new, multi-level prefetching algorithm, we developed prefetching-coordinator (PFC), a hierarchy-aware optimization applicable to any existing prefetching algorithms. PFC does not require any application hints, a priori knowledge on the application access pattern or the native prefetching algorithm, or modification to the I/O interface. Instead, it monitors the upper-level access patterns as well as the lower-level cache status, and dynamically adjusts the aggressiveness of the lower-level prefetching activities. We evaluated PFC with extensive simulation study using a verified multi-level storage simulator, an accurate disk simulator, and access traces with different access patterns. Our results indicate that PFC dynamically controls lower-level prefetching in reaction to multiple system and workload parameters, improving the overall system performance in all 96 test cases. Working with four well-known existing prefetching algorithms adopted in real systems, PFC obtains an improvement of up to 35% to the average request response time, with an average improvement of 14.6% over all cases.

Storage system failure is a serious concern as we approach Petascale computing. Even at today's sub-Petascale levels, I/O failure is the leading cause of downtimes and job failures. We contribute a novel, on-the-fly recovery framework for job input data into supercomputer parallel file systems. The framework exploits key traits ofthe HPC I/O workload to reconstruct lost input data during job execution from remote, immutable copies. Each reconstructed data stripe is made immediately accessible in the client request order due to the delayed metadata update and fine-granular locking while unrelated access to the same file remains unaffected. We have implemented the recovery component within the Lustre parallel file system, thus building a novel application-transparent online recovery solution. Our solution is integrated into Lustre's two-level locking scheme using a two-phase blocking protocol. Combining parametric and simulation studies, our experiments demonstrate a significant improvement in HPC center service ability and user job turnaround time.

Procurement and the optimized utilization of Petascale supercomputers and centers is a renewed national priority. Sustained performance and availability of such large centers is a key technical challenge significantly impacting their usability. Storage systems are known to be the primary fault source leading to data unavailability and job resubmissions. This results in reduced center performance, partially due to the lack of coordination between I/O activities and job scheduling.