HCloud: Resource-efficient provisioning in shared cloud systems

May 26, 2016 ~ Adrian Colyer ~ 2 Comments

HCloud: Resource-efficient provisioning in shared cloud systems - Delimitrou & Kozyrakis, ASPLOS '16 Do you use the public cloud? If so, I'm pretty confident you're going to find today's paper really interesting. Delimitrou & Kozyrakis study the provisioning strategies that provide the best balance between performance and cost. The sweet spot it turns out, is ... Continue Reading

GeePS: Scalable deep learning on distributed GPUs with a GPU-specialized parameter server

April 27, 2016 ~ Adrian Colyer ~ 2 Comments

GeePS: Scalable deep learning on distributed GPUs with a GPU-specialized parameter server - Cui et al. 2016 (EuroSys 2016) We know that deep learning is well suited to GPUs since it has inherent parallelism. But so far this has mostly been limited to either a single GPU (e.g. using Caffe) or to specially built distributed ... Continue Reading

Maglev: A Fast and Reliable Software Network Load Balancer

March 21, 2016 ~ Adrian Colyer ~ 7 Comments

Maglev: A Fast and Reliable Software Network Load Balancer - Eisenbud et al. 2016 Maglev is Google's software load balancer used within all their datacenters. It offers greater scalability and availability than hardware load balancers, enables quick iteration, and is much easier to upgrade. Maglev is a just another distributed system running on the commodity ... Continue Reading

Distributed TensorFlow with MPI

March 18, 2016 ~ Adrian Colyer ~ 4 Comments

Distributed TensorFlow with MPI - Vishnu et al. 2016 A short early release paper to close out the week this week, which looks at how to support machine learning and data mining (MLDM) with Google's TensorFlow in a distributed setting. The paper also contains some good background on TensorFlow itself as well as MPI - ... Continue Reading

Distributed Consistency and Session Anomalies

February 26, 2016 ~ Adrian Colyer ~ 13 Comments

Since we've spent the last couple of days sketching anomaly diagrams and looking at isolation levels, I wanted to finish the week off with a quick recap of session anomalies and consistency levels for distributed stores. In terms of papers, I've drawn primary material for this from: Highly Available Transactions: Virtues and Limitations, and Linearizability ... Continue Reading

The Heard-Of Model: Computing in Distributed Systems with Benign Failures

February 9, 2016 ~ Adrian Colyer ~ Leave a comment

The Heard-Of Model: Computing in Distributed Systems with Benign Failures - Charron-Bost & Schiper2007 We briefly touched on the Heard-Of model last week when we looked at PSync. It's really very elegant, so today I thought it would be good to take a closer look. The traditional view of fault-tolerant distributed systems makes the following ... Continue Reading

Chapar: Certified Causally Consistent Distributed Key-Value Stores

February 8, 2016 ~ Adrian Colyer ~ 1 Comment

Chapar: Certified Causally Consistent Distributed Key-Value Stores - Lesani et al. 2016 Another POPL '16 paper today. The Chapar framework provides for modular verification of causal consistency for both causally consistent key-value store implementations and for client programs written to use them. §1 also wins the prize for best use of emojis in a research ... Continue Reading

‘Cause I’m Strong Enough: Reasoning About Consistency Choices in Distributed Systems

February 3, 2016 ~ Adrian Colyer ~ 3 Comments

'Cause I'm Strong Enough: Reasoning About Consistency Choices in Distributed Systems - Gotsman et al. 2016 With apologies for the longer write-up today, I've tried to stick right to the heart of the matter, but even that takes quite some explanation... We've looked at the theme of coordination avoidance before - instead of uniformly applying ... Continue Reading

PSync: A Partially Synchronous Language for Fault-Tolerant Distributed Algorithms

February 1, 2016 ~ Adrian Colyer ~ 1 Comment

PSync: A Partially Synchronous Language for Fault-Tolerant Distributed Algorithms - Drăgoi et al. 2016 Last month we looked at the RAMCloud team's design pattern for building distributed, concurrent, fault-tolerant modules. Today's paper goes one step beyond a pattern, and introduces a domain-specific language called PSync with the goal of unifying the modeling, programming, and verification ... Continue Reading

Panopticon: An Omniscient Lock Broker for Efficient Distributed Transactions in the Datacenter

January 29, 2016 ~ Adrian Colyer ~ 1 Comment

Panopticon: An Omniscient Lock Broker for Efficient Distributed Transactions in the Datacenter - Tasci & Demirbas, 2015 Today we return to the theme of distributed transactions, and a paper that won a best paper award from IEEE Big Data in 2015. Panopticon is a centralized lock broker (like Chubby and ZooKeeper) that manages distributed (decentralized) ... Continue Reading

the morning paper

a random walk through Computer Science research, by Adrian Colyer
Made delightfully fast by strattic

Distributed Systems