ddia/en-us/ch10.md

# 10. Batch Processing

![](../img/ch10.png)

> *A system cannot be successful if it is too strongly influenced by a single person. Once the initial design is complete and fairly robust, the real test begins as people with many different viewpoints undertake their own experiments.*
>
> ​    — Donald Knuth

---------------

In the first two parts of this book we talked a lot about *requests* and *queries*, and the corresponding *responses* or *results*. This style of data processing is assumed in many modern data systems: you ask for something, or you send an instruction, and some time later the system (hopefully) gives you an answer. Databases, caches, search indexes, web servers, and many other systems work this way.

In such *online* systems, whether it’s a web browser requesting a page or a service call‐ ing a remote API, we generally assume that the request is triggered by a human user, and that the user is waiting for the response. They shouldn’t have to wait too long, so we pay a lot of attention to the *response time* of these systems (see “[Describing Performance](ch1.md#describing-performance)”).

The web, and increasing numbers of HTTP/REST-based APIs, has made the request/ response style of interaction so common that it’s easy to take it for granted. But we should remember that it’s not the only way of building systems, and that other approaches have their merits too. Let’s distinguish three different types of systems:

***Services (online systems)***

A service waits for a request or instruction from a client to arrive. When one is received, the service tries to handle it as quickly as possible and sends a response back. Response time is usually the primary measure of performance of a service, and availability is often very important (if the client can’t reach the service, the user will probably get an error message).

***Batch processing systems (offline systems)***

A batch processing system takes a large amount of input data, runs a *job* to pro‐ cess it, and produces some output data. Jobs often take a while (from a few minutes to several days), so there normally isn’t a user waiting for the job to fin‐ ish. Instead, batch jobs are often scheduled to run periodically (for example, once a day). The primary performance measure of a batch job is usually *throughput* (the time it takes to crunch through an input dataset of a certain size). We dis‐ cuss batch processing in this chapter.

***Stream processing systems (near-real-time systems)***

Stream processing is somewhere between online and offline/batch processing (so it is sometimes called *near-real-time* or *nearline* processing). Like a batch pro‐ cessing system, a stream processor consumes inputs and produces outputs (rather than responding to requests). However, a stream job operates on events shortly after they happen, whereas a batch job operates on a fixed set of input data. This difference allows stream processing systems to have lower latency than the equivalent batch systems. As stream processing builds upon batch process‐ ing, we discuss it in [Chapter 11](ch11.md).

As we shall see in this chapter, batch processing is an important building block in our quest to build reliable, scalable, and maintainable applications. For example, Map‐ Reduce, a batch processing algorithm published in 2004 [1], was (perhaps over- enthusiastically) called “the algorithm that makes Google so massively scalable” [2]. It was subsequently implemented in various open source data systems, including Hadoop, CouchDB, and MongoDB.

MapReduce is a fairly low-level programming model compared to the parallel pro‐ cessing systems that were developed for data warehouses many years previously [3, 4], but it was a major step forward in terms of the scale of processing that could be achieved on commodity hardware. Although the importance of MapReduce is now declining [5], it is still worth understanding, because it provides a clear picture of why and how batch processing is useful.

In fact, batch processing is a very old form of computing. Long before programmable digital computers were invented, punch card tabulating machines—such as the Hol‐ lerith machines used in the 1890 US Census [6]—implemented a semi-mechanized form of batch processing to compute aggregate statistics from large inputs. And Map‐ Reduce bears an uncanny resemblance to the electromechanical IBM card-sorting machines that were widely used for business data processing in the 1940s and 1950s [7]. As usual, history has a tendency of repeating itself.

In this chapter, we will look at MapReduce and several other batch processing algo‐ rithms and frameworks, and explore how they are used in modern data systems. But first, to get started, we will look at data processing using standard Unix tools. Even if you are already familiar with them, a reminder about the Unix philosophy is worthwhile because the ideas and lessons from Unix carry over to large-scale, heterogene‐ ous distributed data systems.



## ……



## Summary


In this chapter we explored the topic of batch processing. We started by looking at Unix tools such as awk, grep, and sort, and we saw how the design philosophy of those tools is carried forward into MapReduce and more recent dataflow engines. Some of those design principles are that inputs are immutable, outputs are intended to become the input to another (as yet unknown) program, and complex problems are solved by composing small tools that “do one thing well.”

In the Unix world, the uniform interface that allows one program to be composed with another is files and pipes; in MapReduce, that interface is a distributed filesys‐ tem. We saw that dataflow engines add their own pipe-like data transport mecha‐ nisms to avoid materializing intermediate state to the distributed filesystem, but the initial input and final output of a job is still usually HDFS.

The two main problems that distributed batch processing frameworks need to solve are:

***Partitioning***

In MapReduce, mappers are partitioned according to input file blocks. The out‐ put of mappers is repartitioned, sorted, and merged into a configurable number of reducer partitions. The purpose of this process is to bring all the related data— e.g., all the records with the same key—together in the same place.

Post-MapReduce dataflow engines try to avoid sorting unless it is required, but they otherwise take a broadly similar approach to partitioning.

***Fault tolerance***

MapReduce frequently writes to disk, which makes it easy to recover from an individual failed task without restarting the entire job but slows down execution in the failure-free case. Dataflow engines perform less materialization of inter‐ mediate state and keep more in memory, which means that they need to recom‐ pute more data if a node fails. Deterministic operators reduce the amount of data that needs to be recomputed.



We discussed several join algorithms for MapReduce, most of which are also inter‐ nally used in MPP databases and dataflow engines. They also provide a good illustra‐ tion of how partitioned algorithms work:

***Sort-merge joins***

Each of the inputs being joined goes through a mapper that extracts the join key. By partitioning, sorting, and merging, all the records with the same key end up going to the same call of the reducer. This function can then output the joined records.

***Broadcast hash joins***

One of the two join inputs is small, so it is not partitioned and it can be entirely loaded into a hash table. Thus, you can start a mapper for each partition of the large join input, load the hash table for the small input into each mapper, and then scan over the large input one record at a time, querying the hash table for each record.

***Partitioned hash joins***

If the two join inputs are partitioned in the same way (using the same key, same hash function, and same number of partitions), then the hash table approach can be used independently for each partition.



Distributed batch processing engines have a deliberately restricted programming model: callback functions (such as mappers and reducers) are assumed to be stateless and to have no externally visible side effects besides their designated output. This restriction allows the framework to hide some of the hard distributed systems prob‐ lems behind its abstraction: in the face of crashes and network issues, tasks can be retried safely, and the output from any failed tasks is discarded. If several tasks for a partition succeed, only one of them actually makes its output visible.

Thanks to the framework, your code in a batch processing job does not need to worry about implementing fault-tolerance mechanisms: the framework can guarantee that the final output of a job is the same as if no faults had occurred, even though in real‐ ity various tasks perhaps had to be retried. These reliable semantics are much stron‐ ger than what you usually have in online services that handle user requests and that write to databases as a side effect of processing a request.

The distinguishing feature of a batch processing job is that it reads some input data and produces some output data, without modifying the input—in other words, the output is derived from the input. Crucially, the input data is *bounded*: it has a known, fixed size (for example, it consists of a set of log files at some point in time, or a snap‐ shot of a database’s contents). Because it is bounded, a job knows when it has finished reading the entire input, and so a job eventually completes when it is done.

In the next chapter, we will turn to stream processing, in which the input is *unboun‐ ded*—that is, you still have a job, but its inputs are never-ending streams of data. In this case, a job is never complete, because at any time there may still be more work coming in. We shall see that stream and batch processing are similar in some respects, but the assumption of unbounded streams also changes a lot about how we build systems.



## References
--------------------

1.  Jeffrey Dean and Sanjay Ghemawat: “[MapReduce: Simplified Data Processing on Large Clusters](http://research.google.com/archive/mapreduce.html),” at *6th USENIX Symposium on Operating System Design and Implementation* (OSDI), December 2004.

1.  Joel Spolsky: “[The Perils of JavaSchools](http://www.joelonsoftware.com/articles/ThePerilsofJavaSchools.html),” *joelonsoftware.com*, December 25, 2005.

1.  Shivnath Babu and Herodotos Herodotou: “[Massively Parallel Databases and MapReduce Systems](http://research.microsoft.com/pubs/206464/db-mr-survey-final.pdf),” *Foundations and Trends in Databases*, volume 5, number 1, pages 1–104, November 2013. [doi:10.1561/1900000036](http://dx.doi.org/10.1561/1900000036)

1.  David J. DeWitt and Michael Stonebraker: “[MapReduce: A Major Step Backwards](https://homes.cs.washington.edu/~billhowe/mapreduce_a_major_step_backwards.html),” originally published at *databasecolumn.vertica.com*, January 17, 2008.

1.  Henry Robinson: “[The Elephant Was a Trojan Horse: On the Death of Map-Reduce at Google](http://the-paper-trail.org/blog/the-elephant-was-a-trojan-horse-on-the-death-of-map-reduce-at-google/),”
    *the-paper-trail.org*, June 25, 2014.

1.  “[The Hollerith Machine](https://www.census.gov/history/www/innovations/technology/the_hollerith_tabulator.html),” United States Census Bureau, *census.gov*.

1.  “[IBM 82, 83, and 84 Sorters Reference Manual](http://www.textfiles.com/bitsavers/pdf/ibm/punchedCard/Sorter/A24-1034-1_82-83-84_sorters.pdf),” Edition A24-1034-1, International Business
    Machines Corporation, July 1962.

1.  Adam Drake: “[Command-Line Tools Can Be 235x Faster than Your Hadoop Cluster](http://aadrake.com/command-line-tools-can-be-235x-faster-than-your-hadoop-cluster.html),” *aadrake.com*, January 25, 2014.

1.  “[GNU Coreutils 8.23 Documentation](http://www.gnu.org/software/coreutils/manual/html_node/index.html),” Free Software Foundation, Inc., 2014.

1.  Martin Kleppmann: “[Kafka, Samza, and the Unix Philosophy of Distributed Data](http://martin.kleppmann.com/2015/08/05/kafka-samza-unix-philosophy-distributed-data.html),” *martin.kleppmann.com*, August 5, 2015.

1.  Doug McIlroy:[Internal Bell Labs memo](http://cm.bell-labs.com/cm/cs/who/dmr/mdmpipe.pdf), October 1964. Cited in: Dennis M. Richie: “[Advice from Doug McIlroy](https://www.bell-labs.com/usr/dmr/www/mdmpipe.html),” *cm.bell-labs.com*.

1.  M. D. McIlroy, E. N. Pinson, and B. A. Tague: “[UNIX Time-Sharing System: Foreword](https://archive.org/details/bstj57-6-1899),” *The Bell System Technical Journal*, volume 57, number 6, pages 1899–1904, July 1978.

1.  Eric S. Raymond: <a href="http://www.catb.org/~esr/writings/taoup/html/">*The Art of UNIX Programming*</a>. Addison-Wesley, 2003. ISBN: 978-0-13-142901-7

1.  Ronald Duncan: “[Text File Formats – ASCII Delimited Text – Not CSV or TAB Delimited Text](https://ronaldduncan.wordpress.com/2009/10/31/text-file-formats-ascii-delimited-text-not-csv-or-tab-delimited-text/),”
    *ronaldduncan.wordpress.com*, October 31, 2009.

1.  Alan Kay: “[Is 'Software Engineering' an Oxymoron?](http://tinlizzie.org/~takashi/IsSoftwareEngineeringAnOxymoron.pdf),” *tinlizzie.org*.

1.  Martin Fowler: “[InversionOfControl](http://martinfowler.com/bliki/InversionOfControl.html),” *martinfowler.com*, June 26, 2005.

1.  Daniel J. Bernstein: “[Two File Descriptors for Sockets](http://cr.yp.to/tcpip/twofd.html),” *cr.yp.to*.

1.  Rob Pike and Dennis M. Ritchie: “[The Styx Architecture for Distributed Systems](http://doc.cat-v.org/inferno/4th_edition/styx),” *Bell Labs Technical Journal*, volume 4, number 2, pages 146–152, April 1999.

1.  Sanjay Ghemawat, Howard Gobioff, and Shun-Tak Leung: “[The Google File System](http://research.google.com/archive/gfs-sosp2003.pdf),” at *19th ACM Symposium on Operating Systems Principles* (SOSP), October 2003. [doi:10.1145/945445.945450](http://dx.doi.org/10.1145/945445.945450)

1.  Michael Ovsiannikov, Silvius Rus, Damian Reeves, et al.: “[The Quantcast File System](http://db.disi.unitn.eu/pages/VLDBProgram/pdf/industry/p808-ovsiannikov.pdf),” *Proceedings of the VLDB Endowment*, volume 6, number 11, pages 1092–1101, August 2013. [doi:10.14778/2536222.2536234](http://dx.doi.org/10.14778/2536222.2536234)

1.  “[OpenStack Swift 2.6.1 Developer Documentation](http://docs.openstack.org/developer/swift/),” OpenStack Foundation, *docs.openstack.org*, March 2016.

1.  Zhe Zhang, Andrew Wang, Kai Zheng, et al.: “[Introduction to HDFS Erasure Coding in Apache Hadoop](http://blog.cloudera.com/blog/2015/09/introduction-to-hdfs-erasure-coding-in-apache-hadoop/),” *blog.cloudera.com*, September 23, 2015.

1.  Peter Cnudde: “[Hadoop Turns 10](http://yahoohadoop.tumblr.com/post/138739227316/hadoop-turns-10),” *yahoohadoop.tumblr.com*, February 5, 2016.

1.  Eric Baldeschwieler: “[Thinking About the HDFS vs. Other Storage Technologies](http://hortonworks.com/blog/thinking-about-the-hdfs-vs-other-storage-technologies/),” *hortonworks.com*, July 25, 2012.

1.  Brendan Gregg: “[Manta: Unix Meets Map Reduce](http://dtrace.org/blogs/brendan/2013/06/25/manta-unix-meets-map-reduce/),” *dtrace.org*, June 25, 2013.

1.  Tom White: *Hadoop: The Definitive Guide*, 4th edition. O'Reilly Media, 2015. ISBN: 978-1-491-90163-2

1.  Jim N. Gray: “[Distributed Computing Economics](http://arxiv.org/pdf/cs/0403019.pdf),” Microsoft Research Tech Report MSR-TR-2003-24, March 2003.

1.  Márton Trencséni: “[Luigi vs Airflow vs Pinball](http://bytepawn.com/luigi-airflow-pinball.html),” *bytepawn.com*, February 6, 2016.

1.  Roshan Sumbaly, Jay Kreps, and Sam Shah: “[The 'Big Data' Ecosystem at LinkedIn](http://www.slideshare.net/s_shah/the-big-data-ecosystem-at-linkedin-23512853),” at *ACM International Conference on Management of Data (SIGMOD)*, July 2013. [doi:10.1145/2463676.2463707](http://dx.doi.org/10.1145/2463676.2463707)

1.  Alan F. Gates, Olga Natkovich, Shubham Chopra, et al.: “[Building a High-Level Dataflow System on Top of Map-Reduce: The Pig Experience](http://www.vldb.org/pvldb/2/vldb09-1074.pdf),” at *35th International Conference on Very Large Data Bases* (VLDB), August 2009.

1.  Ashish Thusoo, Joydeep Sen Sarma, Namit Jain, et al.: “[Hive – A Petabyte Scale Data Warehouse Using Hadoop](http://i.stanford.edu/~ragho/hive-icde2010.pdf),” at *26th IEEE International Conference on Data Engineering* (ICDE), March 2010. [doi:10.1109/ICDE.2010.5447738](http://dx.doi.org/10.1109/ICDE.2010.5447738)

1.  “[Cascading 3.0 User Guide](http://docs.cascading.org/cascading/3.0/userguide/),” Concurrent, Inc., *docs.cascading.org*, January 2016.

1.  “[Apache Crunch User Guide](https://crunch.apache.org/user-guide.html),” Apache Software Foundation, *crunch.apache.org*.

1.  Craig Chambers, Ashish Raniwala, Frances Perry, et al.: “[FlumeJava: Easy, Efficient Data-Parallel Pipelines](https://research.google.com/pubs/archive/35650.pdf),” at *31st ACM SIGPLAN Conference on Programming Language Design and Implementation* (PLDI), June 2010. [doi:10.1145/1806596.1806638](http://dx.doi.org/10.1145/1806596.1806638)

1.  Jay Kreps: “[Why Local State is a Fundamental Primitive in Stream Processing](https://www.oreilly.com/ideas/why-local-state-is-a-fundamental-primitive-in-stream-processing),” *oreilly.com*, July 31, 2014.

1.  Martin Kleppmann: “[Rethinking Caching in Web Apps](http://martin.kleppmann.com/2012/10/01/rethinking-caching-in-web-apps.html),” *martin.kleppmann.com*, October 1, 2012.

1.  Mark Grover, Ted Malaska, Jonathan Seidman, and Gwen Shapira: *[Hadoop Application Architectures](http://shop.oreilly.com/product/0636920033196.do)*. O'Reilly Media, 2015. ISBN: 978-1-491-90004-8

1.  Philippe Ajoux, Nathan Bronson, Sanjeev Kumar, et al.: “[Challenges to Adopting Stronger Consistency at Scale](https://www.usenix.org/system/files/conference/hotos15/hotos15-paper-ajoux.pdf),” at *15th USENIX Workshop on Hot Topics in Operating Systems* (HotOS), May 2015.

1.  Sriranjan Manjunath: “[Skewed Join](https://wiki.apache.org/pig/PigSkewedJoinSpec),” *wiki.apache.org*, 2009.

1.  David J. DeWitt, Jeffrey F. Naughton, Donovan A.Schneider, and S. Seshadri: “[Practical Skew Handling in Parallel Joins](http://www.vldb.org/conf/1992/P027.PDF),” at *18th International Conference on Very Large Data Bases* (VLDB), August 1992.

1.  Marcel Kornacker, Alexander Behm, Victor Bittorf, et al.: “[Impala: A Modern, Open-Source SQL Engine for Hadoop](http://pandis.net/resources/cidr15impala.pdf),” at *7th Biennial Conference on Innovative Data Systems Research* (CIDR), January 2015.

1.  Matthieu Monsch: “[Open-Sourcing PalDB, a Lightweight Companion for Storing Side Data](https://engineering.linkedin.com/blog/2015/10/open-sourcing-paldb--a-lightweight-companion-for-storing-side-da),” *engineering.linkedin.com*, October 26, 2015.

1.  Daniel Peng and Frank Dabek: “[Large-Scale Incremental Processing Using Distributed Transactions and Notifications](https://www.usenix.org/legacy/event/osdi10/tech/full_papers/Peng.pdf),” at *9th USENIX conference on Operating Systems Design and Implementation* (OSDI), October 2010.

1.  “["Cloudera Search User Guide,"](http://www.cloudera.com/documentation/cdh/5-1-x/Search/Cloudera-Search-User-Guide/Cloudera-Search-User-Guide.html) Cloudera, Inc., September 2015.

1.  Lili Wu, Sam Shah, Sean Choi, et al.:
    “[The Browsemaps: Collaborative Filtering at LinkedIn](http://ls13-www.cs.uni-dortmund.de/homepage/rsweb2014/papers/rsweb2014_submission_3.pdf),” at *6th Workshop on Recommender Systems and the Social Web* (RSWeb), October 2014.

1.  Roshan Sumbaly, Jay Kreps, Lei Gao, et al.: “[Serving Large-Scale Batch Computed Data with Project Voldemort](http://static.usenix.org/events/fast12/tech/full_papers/Sumbaly.pdf),” at *10th USENIX Conference on File and Storage Technologies* (FAST), February 2012.

1.  Varun Sharma: “[Open-Sourcing Terrapin: A Serving System for Batch Generated Data](https://engineering.pinterest.com/blog/open-sourcing-terrapin-serving-system-batch-generated-data-0),”  *engineering.pinterest.com*, September 14, 2015.

1.  Nathan Marz: “[ElephantDB](http://www.slideshare.net/nathanmarz/elephantdb),” *slideshare.net*, May 30, 2011.

1.  Jean-Daniel (JD) Cryans: “[How-to: Use HBase Bulk Loading, and Why](http://blog.cloudera.com/blog/2013/09/how-to-use-hbase-bulk-loading-and-why/),” *blog.cloudera.com*, September 27, 2013.

1.  Nathan Marz:  “[How to Beat the CAP   Theorem](http://nathanmarz.com/blog/how-to-beat-the-cap-theorem.html),” *nathanmarz.com*, October 13, 2011.

1.  Molly Bartlett Dishman and Martin Fowler:  “[Agile   Architecture](http://conferences.oreilly.com/software-architecture/sa2015/public/schedule/detail/40388),” at *O'Reilly Software Architecture Conference*, March 2015.

1.  David J. DeWitt and Jim N. Gray: “[Parallel Database Systems: The Future of High Performance Database Systems](http://www.cs.cmu.edu/~pavlo/courses/fall2013/static/papers/dewittgray92.pdf),” *Communications of the ACM*, volume 35, number 6, pages 85–98, June 1992. [doi:10.1145/129888.129894](http://dx.doi.org/10.1145/129888.129894)

1.  Jay Kreps: “[But the multi-tenancy thing is actually really really hard](https://twitter.com/jaykreps/status/528235702480142336),” tweetstorm, *twitter.com*, October 31, 2014.

1.  Jeffrey Cohen, Brian Dolan, Mark Dunlap, et al.: “[MAD Skills: New Analysis Practices for Big Data](http://www.vldb.org/pvldb/2/vldb09-219.pdf),” *Proceedings of the VLDB Endowment*, volume 2, number 2, pages 1481–1492, August 2009. [doi:10.14778/1687553.1687576](http://dx.doi.org/10.14778/1687553.1687576)

1.  Ignacio Terrizzano, Peter Schwarz, Mary Roth, and John E. Colino: “[Data Wrangling: The Challenging Journey from the Wild to the Lake](http://cidrdb.org/cidr2015/Papers/CIDR15_Paper2.pdf),” at *7th Biennial Conference on Innovative Data Systems Research* (CIDR), January 2015.

1.  Paige Roberts: “[To Schema on Read or to Schema on Write, That Is the Hadoop Data Lake Question](http://adaptivesystemsinc.com/blog/to-schema-on-read-or-to-schema-on-write-that-is-the-hadoop-data-lake-question/),” *adaptivesystemsinc.com*, July 2, 2015.

1.  Bobby Johnson and Joseph Adler: “[The Sushi Principle: Raw Data Is Better](https://vimeo.com/123985284),” at *Strata+Hadoop World*, February 2015.

1.  Vinod Kumar Vavilapalli, Arun C. Murthy, Chris Douglas, et al.: “[Apache Hadoop YARN: Yet Another Resource Negotiator](http://www.socc2013.org/home/program/a5-vavilapalli.pdf),” at *4th ACM Symposium on Cloud Computing* (SoCC), October 2013. [doi:10.1145/2523616.2523633](http://dx.doi.org/10.1145/2523616.2523633)

1.  Abhishek Verma, Luis Pedrosa, Madhukar Korupolu, et al.: “[Large-Scale Cluster Management at Google with Borg](http://research.google.com/pubs/pub43438.html),” at *10th European Conference on Computer Systems* (EuroSys), April 2015. [doi:10.1145/2741948.2741964](http://dx.doi.org/10.1145/2741948.2741964)

1.  Malte Schwarzkopf: “[The Evolution of Cluster Scheduler Architectures](http://www.firmament.io/blog/scheduler-architectures.html),” *firmament.io*, March 9, 2016.

1.  Matei Zaharia, Mosharaf Chowdhury, Tathagata Das, et al.: “[Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing](https://www.usenix.org/system/files/conference/nsdi12/nsdi12-final138.pdf),” at *9th USENIX Symposium on Networked Systems Design and Implementation* (NSDI), April 2012.

1.  Holden Karau, Andy Konwinski, Patrick Wendell, and Matei Zaharia: *Learning Spark*. O'Reilly Media, 2015. ISBN: 978-1-449-35904-1

1.  Bikas Saha and Hitesh Shah: “[Apache Tez: Accelerating Hadoop Query Processing](http://www.slideshare.net/Hadoop_Summit/w-1205phall1saha),” at *Hadoop Summit*, June 2014.

1.  Bikas Saha, Hitesh Shah, Siddharth Seth, et al.: “[Apache Tez: A Unifying Framework for Modeling and Building Data Processing Applications](http://home.cse.ust.hk/~weiwa/teaching/Fall15-COMP6611B/reading_list/Tez.pdf),” at *ACM International Conference on Management of Data* (SIGMOD), June 2015. [doi:10.1145/2723372.2742790](http://dx.doi.org/10.1145/2723372.2742790)

1.  Kostas Tzoumas: “[Apache Flink: API, Runtime, and Project Roadmap](http://www.slideshare.net/KostasTzoumas/apache-flink-api-runtime-and-project-roadmap),” *slideshare.net*, January 14, 2015.

1.  Alexander Alexandrov, Rico Bergmann, Stephan Ewen, et al.: “[The Stratosphere Platform for Big Data Analytics](https://ssc.io/pdf/2014-VLDBJ_Stratosphere_Overview.pdf),” *The VLDB Journal*, volume 23, number 6, pages 939–964, May 2014. [doi:10.1007/s00778-014-0357-y](http://dx.doi.org/10.1007/s00778-014-0357-y)

1.  Michael Isard, Mihai Budiu, Yuan Yu, et al.: “[Dryad: Distributed Data-Parallel Programs from Sequential Building Blocks](http://research.microsoft.com/en-us/projects/dryad/eurosys07.pdf),” at *European Conference on Computer Systems* (EuroSys), March 2007. [doi:10.1145/1272996.1273005](http://dx.doi.org/10.1145/1272996.1273005)

1.  Daniel Warneke and Odej Kao: “[Nephele: Efficient Parallel Data Processing in the Cloud](https://stratosphere2.dima.tu-berlin.de/assets/papers/Nephele_09.pdf),” at *2nd Workshop on Many-Task Computing on Grids and Supercomputers* (MTAGS), November 2009. [doi:10.1145/1646468.1646476](http://dx.doi.org/10.1145/1646468.1646476)

1.  Lawrence Page, Sergey Brin, Rajeev Motwani, and Terry Winograd: ["The PageRank"](http://ilpubs.stanford.edu:8090/422/)

1.  Leslie G. Valiant: “[A Bridging Model for Parallel Computation](http://dl.acm.org/citation.cfm?id=79181),” *Communications of the ACM*, volume 33, number 8, pages 103–111, August 1990. [doi:10.1145/79173.79181](http://dx.doi.org/10.1145/79173.79181)

1.  Stephan Ewen, Kostas Tzoumas, Moritz Kaufmann, and Volker Markl: “[Spinning Fast Iterative Data Flows](http://vldb.org/pvldb/vol5/p1268_stephanewen_vldb2012.pdf),” *Proceedings of the VLDB Endowment*, volume 5, number 11, pages 1268-1279, July 2012. [doi:10.14778/2350229.2350245](http://dx.doi.org/10.14778/2350229.2350245)

1.  Grzegorz Malewicz, Matthew H.Austern, Aart J. C. Bik, et al.: “[Pregel: A System for Large-Scale Graph Processing](https://kowshik.github.io/JPregel/pregel_paper.pdf),” at *ACM International Conference on Management of Data* (SIGMOD), June 2010. [doi:10.1145/1807167.1807184](http://dx.doi.org/10.1145/1807167.1807184)

1.  Frank McSherry, Michael Isard, and Derek G. Murray: “[Scalability! But at What COST?](http://www.frankmcsherry.org/assets/COST.pdf),” at *15th USENIX Workshop on Hot Topics in Operating Systems* (HotOS), May 2015.

1.  Ionel Gog, Malte Schwarzkopf, Natacha Crooks, et al.: “[Musketeer: All for One, One for All in Data Processing Systems](http://www.cl.cam.ac.uk/research/srg/netos/camsas/pubs/eurosys15-musketeer.pdf),” at *10th European Conference on Computer Systems* (EuroSys), April 2015. [doi:10.1145/2741948.2741968](http://dx.doi.org/10.1145/2741948.2741968)

1.  Aapo Kyrola, Guy Blelloch, and Carlos Guestrin: “[GraphChi: Large-Scale Graph Computation on Just a PC](https://www.usenix.org/system/files/conference/osdi12/osdi12-final-126.pdf),” at *10th USENIX Symposium on Operating Systems Design and Implementation* (OSDI), October 2012.

1.  Andrew Lenharth, Donald Nguyen, and Keshav Pingali: “[Parallel Graph Analytics](http://cacm.acm.org/magazines/2016/5/201591-parallel-graph-analytics/fulltext),” *Communications of the ACM*, volume 59, number 5, pages 78–87, May [doi:10.1145/2901919](http://dx.doi.org/10.1145/2901919)

1.  Fabian Hüske: “[Peeking into Apache Flink's Engine Room](http://flink.apache.org/news/2015/03/13/peeking-into-Apache-Flinks-Engine-Room.html),” *flink.apache.org*, March 13, 2015.

1.  Mostafa Mokhtar: “[Hive 0.14 Cost Based Optimizer (CBO) Technical Overview](http://hortonworks.com/blog/hive-0-14-cost-based-optimizer-cbo-technical-overview/),” *hortonworks.com*, March 2, 2015.

1.  Michael Armbrust, Reynold S Xin, Cheng Lian, et al.: “[Spark SQL: Relational Data Processing in Spark](http://people.csail.mit.edu/matei/papers/2015/sigmod_spark_sql.pdf),” at *ACM International Conference on Management of Data* (SIGMOD), June 2015. [doi:10.1145/2723372.2742797](http://dx.doi.org/10.1145/2723372.2742797)

1.  Daniel Blazevski: “[Planting Quadtrees for Apache Flink](http://insightdataengineering.com/blog/flink-knn/),” *insightdataengineering.com*, March 25, 2016.

1.  Tom White: “[Genome Analysis Toolkit: Now Using Apache Spark for Data Processing](http://blog.cloudera.com/blog/2016/04/genome-analysis-toolkit-now-using-apache-spark-for-data-processing/),” *blog.cloudera.com*, April 6, 2016.