Accelerate Hadoop MapReduce Performance using Dedicated OrangeFS Servers

By Nicole Hemsoth

September 9, 2013

Recent tests performed at Clemson University achieved a 25 percent improvement in Apache Hadoop Terasort run times by replacing Hadoop Distributed File System (HDFS) with an OrangeFS configuration using dedicated servers. Key components included extension of the MapReduce “FileSystem” class and a Java Native Interface (JNI) shim to the OrangeFS client. No modifications of Hadoop were required, and existing MapReduce jobs require no modification to utilize OrangeFS. The results also demonstrated the ability to deploy MapReduce with a general purpose High Performance file system in a High Performance Computing (HPC) environment, increasing potential for more flexible workflow.

The open source Hadoop MapReduce project has a traditional hardware architecture that differs from standard HPC architecture, where thin clients access remote, shared, and potentially distributed data servers. With HDFS, clients and data servers are paired together, running on the same hardware. Many HPC sites would like to extend their cluster use to support Hadoop MapReduce. With OrangeFS providing distributed storage as part of HPC clusters, they could leverage their existing investment in HPC to run Hadoop MapReduce workloads.

Through testing this configuration, a number of benefits emerged:

  • MapReduce clients accessing a dedicated OrangeFS storage cluster yielded a 25 percent faster combined run time than the traditional approach, where MapReduce clients access data locally for the three operations (teragen, terasort, and teravalidate).
  • OrangeFS and HDFS, without replication enabled, performed similarly under identical local (traditional HDFS) configurations (within 0.2 percent); however, OrangeFS adds the advantages of a general purpose, scale-out file system. With a general purpose file system, applications can read and write data to OrangeFS while it remains available for Hadoop MapReduce job input, improving run time by eliminating time-consuming HDFS stage-in and stage-out operations.
  • Doubling the number of compute nodes accessing the OrangeFS cluster results in ~300 percent improvement on Terasort job run time.
  • OrangeFS provides good results when clients significantly overcommit storage servers.

About OrangeFS

OrangeFS is a user-friendly, open-source, next-generation parallel file system for compute and storage clusters of the future. OrangeFS increases IO performance by storing a file in objects across multiple servers and accessing these objects in parallel. Offering more feature rich data access and manipulation than HDFS, OrangeFS is an ideal tool for storing, processing and analyzing data with MapReduce. A staff of developers support OrangeFS, improving stability and functionality for the base system and developing new interfaces.

OrangeFS has an object-based infrastructure. Each file and directory consists of two or more objects: one primarily containing file metadata, and the other(s) primarily containing file data. Objects may contain both data and metadata as needed to fulfill their role in the file system. This division and distribution of data to the servers is imperceptible to users, who see a traditional, logical file view. The OrangeFS distributed file structure provides outstanding scalability in performance and capacity.

OrangeFS client interfaces work with a range of operating systems, including Linux, Mac OS X and Windows. Compatible client interfaces include Direct Interface, WebDAV, S3, REST, FUSE, Hadoop and MPI-IO.

OrangeFS with Hadoop MapReduce

Hadoop’s abstract FileSystem class allows MapReduce to leverage file systems other than HDFS, with a configuration file that sets the designated file system. Hadoop MapReduce is written in Java, but OrangeFS’s client libraries are written in C. A Java Native Interface (JNI) shim allows data to be passed between programs, avoiding the overhead of memory copies with Java’s NIO Direct ByteBuffer. The JNI shim allows Java code to execute functions present in the OrangeFS Direct Client Interface. The OrangeFS Direct Client Interface Library is a collection of familiar POSIX-like and system standard input/output (stdio.h) library calls designed for parallel access to OrangeFS. OrangeFS differs from HDFS in that it allows modification of data after the initial write.

The Terasort benchmarks successfully explored potential for replacing HDFS with the prerelease version 2.8.8 of OrangeFS, working with the Hadoop 1.x stable release. The two Hadoop configurations which were evaluated are shown in Figure 1.

File System Test Configuration 

Figure 1 Test Configurations

Test Protocols

Hadoop MapReduce File System Test (Figure 2)

To test the impact of replacing HDFS with OrangeFS, developers performed a full terabyte (1 TB) Terasort benchmark on 8 nodes, each running both MapReduce and the file system shown in the first configuration above. The tests were performed on 8 Dell PowerEdge R720s with local SSDs for metadata and 12 2-TB drives for data. In this test, MapReduce ran locally on the same nodes, first over OrangeFS and then over HDFS, interconnected with 10Gb/s Ethernet. Both file systems used the compute nodes for storage as well.

Hadoop MapReduce Remote Client Test

Using the same benchmarks with typical HPC storage architecture, another test, “OFS Remote” in Figure 2, measured how MapReduce performs when data is stored on dedicated, network-connected storage nodes running OrangeFS. Eight additional nodes were used as MapReduce clients, and eight Dell PowerEdge R720s with local SSDs for metadata and 12 2-TB drives for data were used as storage nodes only, shown in the Remote Client Test Configuration in Figure 1.

Results

OrangeFS decreased Terasort run time in the dedicated OrangeFS storage cluster architecture by about 25 percent over the traditional MapReduce architecture, where clients access data from local disks. OrangeFS and HDFS, without replication enabled, performed similarly under identical local (traditional HDFS) configurations (within 0.2 percent); however, OrangeFS adds the advantages of a general purpose, scale-out file system.

Figure 2 Hadoop MapReduce File System Test 

Figure 2 Hadoop MapReduce File System Test

Hadoop MapReduce over OrangeFS with Overcommitted Storage Servers (Figure 3)

A separate test evaluated MapReduce over OrangeFS, overcommitting the storage nodes and evaluating how well this approach scales with more MapReduce clients than storage nodes. The Terasort test was performed with an increasing number of clients utilizing a dedicated OrangeFS cluster composed of 16 Dell PowerEdge R720s with local SSDs for metadata and 12 2-TB drives for data. The Hadoop client nodes had only a single hard disk drive available for intermediate data storage purposes, increasing the time over previous tests where Hadoop clients possessed 12 disks. If the clients used a solid state drive (SSD) for storage and retrieval of intermediate data instead, the slowdown caused by the single disk compared to an array of disks would be alleviated to some extent.

Figure 3 Hadoop MapReduce over OrangeFS with Overcommitted Storage Servers 

Figure 3 Hadoop MapReduce over OrangeFS with Overcommitted Storage Servers

Results

In testing 16, 32, and 64 compute nodes, doubling the number of compute nodes caused a ~300 percent improvement on Terasort job run time. OrangeFS provides good results when clients significantly overcommit the storage servers (4 to 1 in these tests). While providing improvements as a good general purpose file system for MapReduce, OrangeFS is also an excellent concurrent working file system to support the storage needs of other applications while simultaneously serving Hadoop MapReduce.

Benefits

  • OrangeFS enables modification of data anywhere in a file, while HDFS requires copying data before modification, except in the case of Append in the Hadoop 2.x release.
  • OrangeFS replaces the HDFS single namenode with multiple OrangeFS metadata/data servers, reducing task time with improved scalability and eliminating this single point of contention.
  • Potentially, intermediate data can also be written to OrangeFS rather than a temporary folder on each Hadoop client disk, optionally retaining it for use in future jobs and further improving performance with OrangeFS serving the data to MapReduce.

Obstacles

  • Unlike HDFS, OrangeFS doesn’t currently support built-in replication. (OrangeFS can be run in High Availability (HA) mode, and plans for the 3.0 release of OrangeFS include integrated replication for both data and metadata.)
  • OrangeFS and Hadoop are separate installations which must be configured to work together. (Plans for the 2.8.8 release of OrangeFS include a more comprehensive documentation set, including instructions for using Hadoop’s MapReduce with the OrangeFS file system.)

Conclusion

The results demonstrated that replacing HDFS with OrangeFS produced better MapReduce performance for workloads with high volumes of intermediate data, i.e., terasort.

Separating MapReduce clients from storage servers can provide stability in the case of client failure, without the overhead of replication, and eases local disk contention during the reduce stage.

Hadoop MapReduce can leverage OrangeFS as its underlying storage system in an HPC environment. A Portable Batch System (PBS) or Sun Grid Engine (SGE) scheduled HPC environment can support on-demand Hadoop MapReduce clusters deployed and configured automatically, using the open source project “myHadoop.” Researchers could customize a version of myHadoop to schedule on-demand MapReduce clusters, with data persisting on OrangeFS, eliminating HDFS’s time consuming data stage-in and stage-out phases. (myHadoop scripts will be available with the next release of OrangeFS, for running jobs in a scheduled environment.)

Future evaluations may test how performance could be improved, since Hadoop Map and Reduce tasks could be patched to support reading and writing intermediate data to OrangeFS, rather than local disk, improving job run time with faster I/O rates.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

UCSD, AIST Forge Tighter Alliance with AI-Focused MOU

January 18, 2018

The rich history of collaboration between UC San Diego and AIST in Japan is getting richer. The organizations entered into a five-year memorandum of understanding on January 10. The MOU represents the continuation of a 1 Read more…

By Tiffany Trader

New Blueprint for Converging HPC, Big Data

January 18, 2018

After five annual workshops on Big Data and Extreme-Scale Computing (BDEC), a group of international HPC heavyweights including Jack Dongarra (University of Tennessee), Satoshi Matsuoka (Tokyo Institute of Technology), Read more…

By John Russell

Researchers Measure Impact of ‘Meltdown’ and ‘Spectre’ Patches on HPC Workloads

January 17, 2018

Computer scientists from the Center for Computational Research, State University of New York (SUNY), University at Buffalo have examined the effect of Meltdown and Spectre security updates on the performance of popular H Read more…

By Tiffany Trader

HPE Extreme Performance Solutions

HPE and NREL Take Steps to Create a Sustainable, Energy-Efficient Data Center with an H2 Fuel Cell

As enterprises attempt to manage rising volumes of data, unplanned data center outages are becoming more common and more expensive. As the cost of downtime rises, enterprises lose out on productivity and valuable competitive advantage without access to their critical data. Read more…

Fostering Lustre Advancement Through Development and Contributions

January 17, 2018

Six months after organizational changes at Intel's High Performance Data (HPDD) division, most in the Lustre community have shed any initial apprehension around the potential changes that could affect or disrupt Lustre Read more…

By Carlos Aoki Thomaz

UCSD, AIST Forge Tighter Alliance with AI-Focused MOU

January 18, 2018

The rich history of collaboration between UC San Diego and AIST in Japan is getting richer. The organizations entered into a five-year memorandum of understandi Read more…

By Tiffany Trader

New Blueprint for Converging HPC, Big Data

January 18, 2018

After five annual workshops on Big Data and Extreme-Scale Computing (BDEC), a group of international HPC heavyweights including Jack Dongarra (University of Te Read more…

By John Russell

Researchers Measure Impact of ‘Meltdown’ and ‘Spectre’ Patches on HPC Workloads

January 17, 2018

Computer scientists from the Center for Computational Research, State University of New York (SUNY), University at Buffalo have examined the effect of Meltdown Read more…

By Tiffany Trader

Fostering Lustre Advancement Through Development and Contributions

January 17, 2018

Six months after organizational changes at Intel's High Performance Data (HPDD) division, most in the Lustre community have shed any initial apprehension aroun Read more…

By Carlos Aoki Thomaz

When the Chips Are Down

January 11, 2018

In the last article, "The High Stakes Semiconductor Game that Drives HPC Diversity," I alluded to the challenges facing the semiconductor industry and how that may impact the evolution of HPC systems over the next few years. I thought I’d lift the covers a little and look at some of the commercial challenges that impact the component technology we use in HPC. Read more…

By Dairsie Latimer

How Meltdown and Spectre Patches Will Affect HPC Workloads

January 10, 2018

There have been claims that the fixes for the Meltdown and Spectre security vulnerabilities, named the KPTI (aka KAISER) patches, are going to affect applicatio Read more…

By Rosemary Francis

Momentum Builds for US Exascale

January 9, 2018

2018 looks to be a great year for the U.S. exascale program. The last several months of 2017 revealed a number of important developments that help put the U.S. Read more…

By Alex R. Larzelere

ANL’s Rick Stevens on CANDLE, ARM, Quantum, and More

January 8, 2018

Late last year HPCwire caught up with Rick Stevens, associate laboratory director for computing, environment and life Sciences at Argonne National Laboratory, f Read more…

By John Russell

Inventor Claims to Have Solved Floating Point Error Problem

January 17, 2018

"The decades-old floating point error problem has been solved," proclaims a press release from inventor Alan Jorgensen. The computer scientist has filed for and Read more…

By Tiffany Trader

US Coalesces Plans for First Exascale Supercomputer: Aurora in 2021

September 27, 2017

At the Advanced Scientific Computing Advisory Committee (ASCAC) meeting, in Arlington, Va., yesterday (Sept. 26), it was revealed that the "Aurora" supercompute Read more…

By Tiffany Trader

Japan Unveils Quantum Neural Network

November 22, 2017

The U.S. and China are leading the race toward productive quantum computing, but it's early enough that ultimate leadership is still something of an open questi Read more…

By Tiffany Trader

AMD Showcases Growing Portfolio of EPYC and Radeon-based Systems at SC17

November 13, 2017

AMD’s charge back into HPC and the datacenter is on full display at SC17. Having launched the EPYC processor line in June along with its MI25 GPU the focus he Read more…

By John Russell

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

IBM Begins Power9 Rollout with Backing from DOE, Google

December 6, 2017

After over a year of buildup, IBM is unveiling its first Power9 system based on the same architecture as the Department of Energy CORAL supercomputers, Summit a Read more…

By Tiffany Trader

Fast Forward: Five HPC Predictions for 2018

December 21, 2017

What’s on your list of high (and low) lights for 2017? Volta 100’s arrival on the heels of the P100? Appearance, albeit late in the year, of IBM’s Power9? Read more…

By John Russell

Chip Flaws ‘Meltdown’ and ‘Spectre’ Loom Large

January 4, 2018

The HPC and wider tech community have been abuzz this week over the discovery of critical design flaws that impact virtually all contemporary microprocessors. T Read more…

By Tiffany Trader

Leading Solution Providers

Perspective: What Really Happened at SC17?

November 22, 2017

SC is over. Now comes the myriad of follow-ups. Inboxes are filled with templated emails from vendors and other exhibitors hoping to win a place in the post-SC thinking of booth visitors. Attendees of tutorials, workshops and other technical sessions will be inundated with requests for feedback. Read more…

By Andrew Jones

Researchers Measure Impact of ‘Meltdown’ and ‘Spectre’ Patches on HPC Workloads

January 17, 2018

Computer scientists from the Center for Computational Research, State University of New York (SUNY), University at Buffalo have examined the effect of Meltdown Read more…

By Tiffany Trader

Tensors Come of Age: Why the AI Revolution Will Help HPC

November 13, 2017

Thirty years ago, parallel computing was coming of age. A bitter battle began between stalwart vector computing supporters and advocates of various approaches to parallel computing. IBM skeptic Alan Karp, reacting to announcements of nCUBE’s 1024-microprocessor system and Thinking Machines’ 65,536-element array, made a public $100 wager that no one could get a parallel speedup of over 200 on real HPC workloads. Read more…

By John Gustafson & Lenore Mullin

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Flipping the Flops and Reading the Top500 Tea Leaves

November 13, 2017

The 50th edition of the Top500 list, the biannual publication of the world’s fastest supercomputers based on public Linpack benchmarking results, was released Read more…

By Tiffany Trader

How Meltdown and Spectre Patches Will Affect HPC Workloads

January 10, 2018

There have been claims that the fixes for the Meltdown and Spectre security vulnerabilities, named the KPTI (aka KAISER) patches, are going to affect applicatio Read more…

By Rosemary Francis

GlobalFoundries, Ayar Labs Team Up to Commercialize Optical I/O

December 4, 2017

GlobalFoundries (GF) and Ayar Labs, a startup focused on using light, instead of electricity, to transfer data between chips, today announced they've entered in Read more…

By Tiffany Trader

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

  • arrow
  • Click Here for More Headlines
  • arrow
Share This