| Evaluation seq. num.
|| For the first evaluation leave this field at "1"
|| Unique ID of the evaluator that carried out this specific evaluator.
|Evaluation describtion||text|| The workflow has been implemented as a native JAVA map/reduce application. It uses the Apache Tika™ 1.0 API (detector call) to detect the MIME type of the inputStream for each file inside the ARC.GZ container files.
To run over all items inside the ARC.GZ files, the native JAVA map/reduce program uses a custom RecordReader based on the Hadoop 0.20 API. The custom RecordReader enables the program to read the ARC.GZ files natively and iterate over the archive file record by record (content file by content file). Each record is processed by a single map method call to detect its MIME type.
Each test ARC.GZ file has a size of approximately 500MB and is the container for around 30000 files.
The 100GB test sample (200 x 500MB) is a subset of the original data set produced by a web crawler at the ONB.
Output of the map/reduce program is a MIME type distribution list of the analyzed input, containing all identified MIME types plus the occurrence count for each identified MIME type.
Goal / Sub-goal:
Performance efficiency / Throughput
) The result has been measured as GB/min/platform
Reliability / Stability Indicators
) The processing application has been implemented as a JAVA JAR map / reduce application
) All needed components (program logic, Hadoop method implementations, dependencies, Apache Tika™ 1.0 JAR) are integrated
) The result has been measured "manually" and reflected as a boolean value (true = met the requirements)
Reliability / Runtime stability
) Use Hadoop admin interface to identify failed tasks.
) Use Hadoops output to identify dropped records / any reported errors.
) The result has been measured as an integer value reflecting the number of identified run time failures.
| Textual description of the evaluation and the overall goals
|| Date of evaluation
||Platform ONB 1|| Unique ID of the platform involved in the particular evaluation - see Platform page included below
||100GB sub set of Austrian National Library - Web Archive|| Link to dataset page(s) on WIKI
|Workflow method|| string
||Hadoop map / reduce application implemented in JAVA (jar).|| Taverna / Commandline / Direct hadoop etc...
| Workflow(s) involved
|| Link(s) to MyExperiment if applicable
| Tool(s) involved
||URL(s)||Hadoop cluster, tb-wc-hd-archd, Apache Tika™ 1.0 API|| Link(s) to distinct versions of specific components/tools in the component registry if applicable
|Link(s) to Scenario(s)|| URL(s)
||http://wiki.opf-labs.org/display/SP/WCT4+Web+Archive+Mime-Type+detection+at+Austrian+National+Library|| Link(s) to scenario(s) if applicable
|Platform-ID||String|| ONB 1
|| Unique string that identifies this specific platform.
Use the platform name
|Platform description||String|| Experimental cluster (setup 06.2012).
8 (HT) cores per node. Using max. 7 cores for map / reduce slots (one for the OS).
Map / reduce slots ratio 6 / 1.
|Human readable description of the platform. Where is it located, contact info, etc.|
|Number of nodes||integer||5|| Number of hosts involved - could be both physical hosts as well as virtual hosts
|Total number of physical CPUs||integer||5|| Number of CPU's involved
|CPU specs||string||Xeon X3440@2.53GHz Quadcore CPU|| Specification of CPUs
|Total number of CPU-cores||integer|| 40 Cores (5 * 8 Cores)
|| Number of CPU-cores involved (4 physical Cores + 4 HT Cores = 8 Cores)
| Total amount of RAM in Gbytes
||integer|| 80GB (5 * 16GB)
|| Total amount of RAM on each nodes
| average CPU-cores for nodes
||integer|| 8 Cores
|| Number of CPU-cores in average across all nodes
| avarage RAM in Gbytes for nodes
||integer|| 16 GB
|| Amount of memory in average across all nodes
| Operating System on nodes
||String|| Ubuntu 10.04.04 LTS (64bit)
||Linux (specific distribution), Windows (specific distribution), other?|
|Storage system/layer||String|| HDFS
||NFS, HDFS, local files, ?|
| Disk subsystem
||2 x 1TB DISKs; configured as RAID0 => 2TB effective disk space|| Disk subsystem on each node
| HDFS replication factor
|Network layer between nodes||String||The CONTROLLER and the NODEs are connected to a GBit high performance network switch (guarantees the full GBit performance for each port).||Speed of network interfaces, general network speed|
|Controller: CPU specs||String|| 2 x Xeon E5620@2.40GHz Quadcore CPU
|Controller: RAM|| integer
|| 24 GB
|Controller: Disk subsystem||String||3 x 1TB DISKs; configured as RAID5 => 2TB effective disk space||
metrics must come from / be registered in the metrics catalogue
|Metric||Baseline definition||Baseline value||Goal|| Evaluation 1 (28/08/12)
|| Evaluation 2 (date)
|| Evaluation 3 (date)
|ThroughputGbytesPerMinute||Virtual machine, Ubuntu Linux, 2GB RAM, Core i5 2,5GHz (single Processor VM configuration), Taverna Workbench workflow, TIKA 0.7 in API mode.||0,08||5||16,17|
|ReliableAndStableAssessment||The workflow incorporates different technologies (script, jar, beanshell, Taverna, unix tools) which makes it hard(er) to implement a reliable error handling (compared to a Java map/reduce implementation).|| false
|NumberOfFailedFiles|| n/a (much smaller data set)