August 16, 2012

Workload Analysis of a Large-Scale Key-Value Store

ACM International Conference on Measurement and Modeling of Computer Systems (SIGMETRICS)

Key-value stores are a vital component in many scale-out enterprises, including social networks, online retail, and risk analysis. Accordingly, they are receiving increased attention from the research community in an effort to improve their performance, scalability, reliability, cost, and power consumption. To be effective, such efforts require a detailed understanding of realistic key-value workloads. And yet little is known about these workloads outside of the companies that operate them. This paper aims to address this gap.

Berk Atikoglu, Yuehai Xu, Eitan Frachtenberg, Song Jiang, Michael Paleczny
August 13, 2012

DeTail: Reducing the Flow Completion Time Tail in Datacenter Networks

ACM Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications (SIGCOMM)

Web applications have now become so sophisticated that rendering a typical page may require hundreds of intra-datacenter flows. At the same time, web sites must meet strict page creation deadlines of 200-300ms to satisfy user demands for interactivity. Long-tailed flow completion times make it challenging for web sites to meet these constraints. They are forced to choose between rendering a subset of the complex page, or delay its rendering, thus missing deadlines and sacrificing either quality or responsiveness. Either option leads to potential financial loss.

Dhruba Borthakur, Randy Katz, Prashanth Mohan, Tathagata Das, David Zats
August 12, 2012

Active Sampling for Entity Matching

ACM Conference on Knowledge Discovery and Data Mining (KDD)

In entity matching, a fundamental issue while training a classifier to label pairs of entities as either duplicates or non-duplicates is the one of selecting informative examples. Although active learning presents an attractive solution to this problem, previous approaches minimize the misclassification rate (0-1 loss) of the classifier, which is an unsuitable metric for entity matching due to class imbalance (i.e., many more non-duplicate pairs than duplicate pairs).

Kedar Bellare, Suresh Iyengar Parthasarathy, Aditya Parameswaran, Vibhor Rastogi
June 17, 2012

Storage Infrastructure Behind Facebook Messages: Using HBase at Scale

IEEE International Conference on Data Engineering (ICDE)

Facebook Messages, which combines messages, chat and email into a real-time conversation, is the first application in Facebook to use HBase in production.

Amitanand Aiyer, Mikhail Bautin, Guoqiang Jerry Chen, Pritam Damania, Prakash Khemani, Kannan Muthukkaruppan, Karthik Ranganathan, Nicolas Spiegelberg, Liyin Tang, Madhuwanti Vaidya
June 1, 2012

Power and Performance Evaluation of Memcached on the TILEPro64 Architecture

Sustainable Computing: Informatics and Systems

Power consumption of data centers had become an important factor in the economy and sustainability of large-scale Web services. Researchers and practitioners are spending considerable effort to characterize Web-scale workloads and evaluating their applicability to alternative, more power-efficient architectures. One such workload in particular is the caching layer, which stores expensive-to-regenerate data in fast storage to reduce service times.

Mateusz Berezecki, Eitan Frachtenberg, Michael Paleczny, Ken Steele
May 1, 2012

Thermal Design in the Open Compute Datacenter

IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (ITHERM)

The advent of Web-based services and cloud computing has instigated an explosive growth in demand for datacenters. Traditionally, Internet companies would lease datacenter space and servers from vendors that often emphasize flexibility over efficiency. But as these companies grew larger, they sought to reduce acquisition and operation costs by building their own datacenters.

Eitan Frachtenberg, Dan Lee, Marco Magarelli, Veerendra Mulay, Jay Park
April 25, 2012

PACMan: Coordinated Memory Caching for Parallel Jobs

USENIX Symposium on Networked Systems Design and Implementation (NSDI)

Data-intensive analytics on large clusters is important for modern Internet services. As machines in these clusters have large memories, in-memory caching of inputs is an effective way to speed up the…

Ganesh Ananthanarayanan, Ali Ghodsi, Andrew Wang, Dhruba Borthakur, Srikanth Kandula, Scott Shenker, Ion Stoica
March 1, 2012

Predicting Memcache Throughput using Simulation and Modeling

IEEE Symposium on Theory of Modeling and Simulation (TMS)

The current work introduces a method for predicting Memcached throughput on single-core and multi-core processors. The method is based on traces collected from a full system simulator running Memcached.

Steven Hart, Eitan Frachtenberg, Mateusz Berezecki
December 23, 2011

High-efficiency server design

ACM Conference on Supercomputing (ICS)

Large-scale datacenters consume megawatts in power and cost hundreds of millions of dollars to equip. Reducing the energy and cost footprint of servers can therefore have substantial impact.

Eitan Frachtenberg, Ali Heydari, Hu Li, Amir Michael, Jacob Na, Avery Nisbet, Pierluigi Sarti
July 1, 2011

Many-core key-value store

International Green Computing Conference (IGCC)

Scaling data centers to handle task-parallel workloads requires balancing the cost of hardware, operations, and power. Low-power, low-core-count servers reduce costs in one of these dimensions, but may require additional nodes to provide the required quality of service or increase costs by underutilizing memory and other resources.

Mateusz Berezecki, Eitan Frachtenberg, Michael Paleczny, Ken Steele
June 20, 2011

YSmart: Yet Another SQL-to-MapReduce Translator

International Conference on Distributed Computing Systems (ICDCS)

MapReduce has become an effective approach to big data analytics in large cluster systems, where SQL-like queries play important roles to interface between users and systems. However, based on our Face book daily operation results, certain types of queries are executed at an unacceptable low speed by Hive (a production SQL-to-MapReduce translator). In this paper, we demonstrate that existing SQL-to-MapReduce translators that operate in a one-operation-to-one-job mode and do not consider query correlations cannot generate high-performance MapReduce programs for certain queries, due to the mismatch between complex SQL structures and simple MapReduce framework. We propose and develop a system called Y Smart, a correlation aware SQL-to-MapReduce translator. Y Smart applies a set of rules to use the minimal number of MapReduce jobs to execute multiple correlated operations in a complex query. Y Smart can significantly reduce redundant computations, I/O operations and network transfers compared to existing translators. We have implemented Y Smart with intensive evaluation for complex queries on two Amazon EC2 clusters and one Face book production cluster. The results show that Y Smart can outperform Hive and Pig, two widely used SQL-to-MapReduce translators, by more than four times for query execution.

Rubao Lee, Tian Luo, Yin Huai, Fusheng Wang, Yongqiang He, Xiaodong Zhang
June 12, 2011

Apache Hadoop goes realtime at Facebook

ACM Special Interest Group on Management of Data (SIGMOD)

Facebook recently deployed Facebook Messages, its first ever user-facing application built on the Apache Hadoop platform. Apache HBase is a database-like layer built on Hadoop designed to support billions of messages per day.

Dhruba Borthakur, Joydeep Sen Sarma, Jonathan Gray, Kannan Muthukkaruppan, Nicolas Spiegelberg, Hairong Kuang, Karthik Ranganathan, Dmytro Molkov, Aravind Menon, Sam Rash, Rodrigo Schmidt, Amitanand Aiyer
April 10, 2011

Facebook Immune System

Workshop on Social Network Systems (SNS)

Popular Internet sites are under attack all the time from phishers, fraudsters, and spammers. They aim to steal user information and expose users to unwanted spam. The attackers have vast resources at their disposal. They are well-funded, with full-time skilled labor, control over compromised and infected accounts, and access to global botnets.

Tao Stein, Roger Chen, Karan Mangla
March 30, 2011

FATE and DESTINI: A Framework for Cloud Recovery Testing

USENIX Symposium on Networked Systems Design and Implementation (NSDI)

As the cloud era begins and failures become commonplace, the fate and destiny of availability, reliability and performance are in the hands of failure recovery. Unfortunately, recovery problems still take place, causing downtimes, data loss, and many other problems.

Haryadi S. Gunawi, Thanh Do, Pallavi Joshi, Peter Alvaro, Joseph M. Hellerstein, Andrea C. Arpaci-Dusseau, Remzi H. Arpaci-Dusseau, Koushik Sen, Dhruba Borthakur
January 1, 2011

RCFile: A Fast and Space-efficient Data Placement Structure in MapReduce-based Warehouse Systems

IEEE International Conference on Data Engineering (ICDE)

MapReduce-based data warehouse systems are playing important roles of supporting big data analytics to understand quickly the dynamics of user behavior trends and their needs in typical Web service providers and social network sites (e.g., Facebook). In such a system, the data placement structure is a critical factor that can affect the warehouse performance in a fundamental way.

Yongqiang He, Rubao Lee, Yin Huai, Zheng Shao, Namit Jain, Xiaodong Zhang, Zhiwei Xu
October 4, 2010

Finding a needle in Haystack: Facebook’s photo storage

USENIX Symposium on Operating Systems Design and Implementation (OSDI)

This paper describes Haystack, an object storage system optimized for Facebook’s Photos application. Facebook currently stores over 260 billion images, which translates to over 20 petabytes of data. U…

Doug Beaver, Sanjeev Kumar, Harry Li, Jason Sobel, Peter Vajgel
June 6, 2010

Data warehousing and analytics infrastructure at Facebook.

Special Interest Group on Management of Data (SIGMOD)

Scalable analysis on large data sets has been core to the functions of a number of teams at Facebook – both engineering and non-engineering. Apart from ad hoc analysis of data and creation of business intelligence dashboards by analysts across the company, a number of Facebook’s site features are also based on analyzing large data sets.

Ashish Thusoo, Dhruba Borthakur, Raghotham Murthy, Zheng Shao, Namit Jain, Hao Liu, Suresh Antony, Joydeep Sen Sarma
April 13, 2010

Job Scheduling for Multi-User MapReduce Clusters

ACM European Conference on Computer Systems (EUROSYS)

Sharing a MapReduce cluster between users is attractive because it enables statistical multiplexing (lowering costs) and allows users to share a common large data set. However, we find that traditiona…

Matei Zaharia, Dhruba Borthakur, Joydeep Sen Sarma, Khaled Elmeleegy, Scott Shenker, Ion Stoica
August 1, 2009

Hive – A Warehousing Solution Over a Map-Reduce Framework

International Conference on Very Large Data Bases (VLDB)

The size of data sets being collected and analyzed in the industry for business intelligence is growing rapidly, making traditional warehousing solutions prohibitively expensive. Hadoop is a popular o…

Ashish Thusoo, Joydeep Sen Sarma, Namit Jain, Zheng Shao, Prasad Chakka, Suresh Antony, Hao Liu, Pete Wyckoff