What is Twitter

We all know that doctors make the worst patients because they always laughingly tell us so. But how do we rate at taking our own medicine?

view from the chair: I go to a friend

NAND flash-based solid-state drives (SSDs) are increasingly being deployed in storage systems at different levels such as buffer-caches and even secondary storage. However, the poor reliability and performance offered by these SSDs for write-intensive workloads continues to be their key shortcoming. Several solutions based on traditionally popular notions of temporal and spatial locality help reduce write traffic for SSDs. However, another form of locality - value locality - has remained completely unexplored. Value locality implies that certain data items (i.e., "values," not just logical addresses) are likely to be accessed preferentially. Given evidence for the presence of significant value locality in real-world workloads, we design CA-SSD which employs content-addressable storage (CAS) to exploit such locality. Our CA-SSD design employs enhancements primarily in the flash translation layer (FTL) with minimal additional hardware, suggesting its feasibility. Using three real-world workloads with content information, we devise statistical characterizations of two aspects of value locality - value popularity and temporal value locality - that form the foundation of CA-SSD. We observe that CA-SSD is able to reduce average response times by about 59-84% compared to traditional SSDs. Even for workloads with little or no value locality, CA-SSD continues to offer comparable performance to a traditional SSD. Our findings advocate adoption of CAS in SSDs, paving the way for a new generation of these devices.

/pdf/leveraging-value-locality-in-optimizing-nand-flash-based-450di6xnk6.pdf

Leveraging value locality in optimizing NAND flash-based SSDs

This paper presents an erasure-coded Byzantine fault-tolerant block storage protocol that is nearly as efficient as protocols that tolerate only crashes. Previous Byzantine fault-tolerant block storage protocols have either relied upon replication, which is inefficient for large blocks of data when tolerating multiple faults, or a combination of additional servers, extra computation, and versioned storage. To avoid these expensive techniques, our protocol employs novel mechanisms to optimize for the common case when faults and concurrency are rare. In the common case, a write operation completes in two rounds of communication and a read completes in one round. The protocol requires a short checksum comprised of cryptographic hashes and homomorphic fingerprints. It achieves throughput within 10% of the crash-tolerant protocol for writes and reads in failure-free runs when configured to tolerate up to 6 faulty servers and any number of faulty clients.

http://sosp2007.org/papers/sosp087-hendricks.pdf

Low-overhead byzantine fault-tolerant storage

The article reviews the book "The Innovator's Dilemma: The Revolutionary Book That Will Change the Way You Do Business" by Clayton M. Christensen.

The Innovator's Dilemma: The Revolutionary Book That Will Change the Way You Do Business.

Content platform management is enhanced by logically partitioning a physical cluster that comprises a redundant array of independent nodes. Using an interface, an administrator defines one or more "tenants" within the archive cluster, wherein a tenant has a set of attributes including, for example, namespaces, administrative accounts, data access accounts, and a permission mask. A namespace is a logical partition of the cluster that serves as a collection of objects typically associated with at least one defined application. Each namespace has a private file system such that access to one namespace (and its associated objects) does not enable a user to access objects in another namespace. A namespace has capabilities (e.g., read, write, delete, purge, and the like) that a namespace administrator can choose to enable or disable for a given data account. Using the interface, an administrator for the tenant creates and manages namespaces such that the cluster then is logically partitioned into a set of namespaces, wherein one or more namespaces are associated with a given tenant. This approach enables a user to segregate cluster data into logical partitions. Using the administrative interface, a namespace associated with a given tenant is selectively configured without affecting a configuration of at least one other namespace in the set of namespaces. This architecture enables support for many top level tenants, with multiple namespaces per tenant, and wherein configuration is effected at the level of a namespace.

Fixed content storage within a partitioned content platform using namespaces

Centera uses cryptographic hash functions as a means of addressing stored objects, thus creating a new class of data storage referred to as CAS (content addressed storage). Such hashing serves the useful function of providing a means of uniquely identifying data and providing a global handle to that data, referred to as the Content Address or CA. However, such a model begs the question: how certain can one be that a given CA is indeed unique? In this paper we describe fundamental concepts of cryptographic hash functions, such as collision resistance, preimage resistance, and second-preimage resistance. We then map these properties to the MD5 and SHA-256 hash algorithms, which are used to generate the Centera content address. Finally, we present a proof of the collision resistance of the Centera Content Address.

Collision and Preimage Resistance of the Centera Content Address

In this paper we look at the growth of distributed object stores (DOS) and examine the underlying mechanisms that guide their use and development. Our focus is on the fundamental principles of operation that define this class of system, how it has evolved, and where it is heading as new markets expand beyond the use originally presented. We conclude by speculating about how object stores as a class must evolve to meet the more demanding requirements of future applications.

Distributed Object Store Principles of Operation The Case for Intelligent Storage.

In and of itself, data storage has apparent business utility. But when we can convert data to information, the utility of stored data increases dramatically. It is the layering of relation atop the data mass that is the engine for such conversion. Frank relation amongst discrete objects sporadically ingested is rare, making the process of synthesizing such relation all the more challenging, but the challenge must be met if we are ever to see an equivalent business value for unstructured data as we already have with structured data. This paper describes a novel construct, referred to as a relational distributed object store (RDOS), that seeks to solve the twin problems of how to persistently and reliably store petabytes of unstructured data while simultaneously creating and persisting relations amongst billions of objects.

Creating a Relational Distributed Object Store (特集 情報活用が加速する社会とビジネスのイノベーション)

/pdf/creating-a-relational-distributed-object-store-3db86w0cd8.pdf

Robert Primmer

Papers

Fixed content storage within a partitioned content platform using namespaces

Collision and Preimage Resistance of the Centera Content Address

Distributed Object Store Principles of Operation The Case for Intelligent Storage.

Creating a Relational Distributed Object Store (特集情報活用が加速する社会とビジネスのイノベーション)

Creating a relational distributed object store

Robert Primmer

Papers

Fixed content storage within a partitioned content platform using namespaces

Collision and Preimage Resistance of the Centera Content Address

Distributed Object Store Principles of Operation The Case for Intelligent Storage.

Creating a Relational Distributed Object Store (特集 情報活用が加速する社会とビジネスのイノベーション)

Creating a relational distributed object store

Creating a Relational Distributed Object Store (特集情報活用が加速する社会とビジネスのイノベーション)