Availability in Globally Distributed Storage Systems

By D. Ford, François Labelle, Florentina I. Popovici, Murray Stokely, Van-Anh Truong, Luiz André Barroso, Carrie Grimes, Sean Quinlan. USENIX Symposium on Operating Systems Design and Implementation, 2010

741 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/a090e4f09ab702c21b3e71eeda265550f59f997a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Centralized Repair of Multiple Node Failures With Applications to Communication Efficient Secret Sharing (2016, unknown): “In [9], Ford et al.”
Understanding network failures in data centers: measurement, analysis, and implications (2011, background): “Note that logged events do not necessarily result in a service outage because of failure-mitigation techniques such as network redundancy [1] and replication of compute and data [11,27], typically deployed in data centers.”
Analyzing the download time of availability codes (2015, methodology): “, Amazon’s Dynamo [13], Facebook storage [48]; (ii) (9, 6)-MDS code, which is used in Google file system [16]; (iii) (10, 6, 2, 1)-LRC, which is used in Windows Azure Storage [24]; and (iv) (7, 3, 2, 3)-availability code, which is constructed using the binary (7, 3) simplex code2.”
Erasure Coding in Windows Azure Storage (2012, methodology): “In our study, we focus on independent failures, but the study can be readily generalized to correlated failures [10, 34].”
A Survey on the Edge Computing for the Internet of Things (2018, methodology): “Based on empirical and statistical results, less than 10% of failures last longer than 15minutes [82].”
One extra bit of download ensures perfectly private information retrieval (2014, background): “This solution is clearly impractical due to the massive amount of download required, but is the only feasible solution if there is only a single storage node storing the data [5].”
A Solution to the Network Challenges of Data Recovery in Erasure-coded Distributed Storage Systems: A Study on the Facebook Warehouse Cluster (2013, background): “There have been several studies on failure statistics in storage systems, e.g., see [6, 7] and the references therein.”
Automatic management of partitioned, replicated search services (2011, background): “After recovering from machine failures (which are relatively brief for the most part [8]), an index server will typically rejoin the same ServerSet (since it will be under-replicated) and continue serving requests—all without manual intervention.”
Batch codes through dense graphs without short cycles (2014, background): “c events may correspond to the permanent or persistent loss of a large number of servers due to multiple reasons, including hardware failures, power outages, and loss of rack switches in data centers [9, 12]. Therefore, it is desirable to work with codes that have large minimum 10If k1+is not an int”
Task Failure Prediction in Cloud Data Centers Using Deep Learning (2019, background): “[22] studied the impact of correlated failures on availability of distributed storage systems for Google clusters.”