View Source

h2. Investigator(s)

Sven Schlarb

h2. Dataset

[SP:Austrian National Library Tresor Music Collection]



h2. Platform

[SP:ONB Hadoop Platform]



h2. Purpose of this experiment

The purpose of this experiment is to evaluate the performance of a scalable workflow for migrating TIFF images to images in the JPEG2000 format compared to an equivalent Taverna version of the workflow processing the data sequentially.

h2. Evaluation method

A Taverna workflow for sequential processing serves as a reference point for the large-scale execution. Out of the full Austrian National Library Tresor Music Collection data set subsets of increasing size are selected by a random process.

The following bash statement is used to create a random sample from the full data set:

{code}
find . -type f -exec ls -l -sd {} + | grep ".tif" | \
awk 'BEGIN {srand()} {printf "%05.0f %s \n",rand()*99999, $0; }' | \
sort -n |  awk '{print $10 "\t" $7}' | head -$NUM > ~/tresormusicfilepaths${NUM}_withsize.csv
{code}
The statement prepends a random number to the file paths list and orders the list subsequently. Variable NUM is the desired size of the data set. The resulting file contains the local file paths and can be used as input for the Taverna workflow presented in the next section.

Additionally the files are uploaded to HDFS as input for the large-scale workflow execution.

By that way it is possible to compare the sequential execution time to the large-scale processing time.

h2. Taverna workflow - sequential processing

The proof-of-concept version of the TIFF to JPEG2000 image migration workflow with quality assurance was created as a Taverna workflow illustrated by the following workflow diagram:

!TavernaWorkflow4276.png|border=1,width=235,height=786!

_Figure 1 (above): Taverna workflow_


Diagram of the TIFF to JPEG2000 image migration workflow, Workflow available on MyExperiment at [http://www.myexperiment.org/workflows/4276.html]
The Taverna workflow reads a textfile containing absolute paths to TIF image files and converts them to JP2 image files using OpenJPEG ([https://code.google.com/p/openjpeg|https://code.google.com/p/openjpeg]).

Based on the input text file, the workflow creates a Taverna list to be processed file by file. A temporary directory is created (createtmpdir) where the migrated image files and some temporary tool outputs are stored.

Before starting the actual migration, it is checked if the TIF input images are valid file format instances using Fits ([https://code.google.com/p/fits|https://code.google.com/p/fits], JHove2 under the hood, [http://www.jhove2.org|http://www.jhove2.org]). An XPath service is used to extract the validity information from the XML-based Fits validation report.

If the images are valid TIF images, they are migrated to the JPEG2000 (JP2) image file format using OpenJPEG 2.0 (opj_compress).

Subsequently, it is again checked if the migrated images are valid JP2 images using SCAPE tool Jpylyzer ([http://www.openplanetsfoundation.org/software/jpylyzer|http://www.openplanetsfoundation.org/software/jpylyzer]). An XPath service (XPathJpylyzer) is used to extract the validity information from the XML-based Jpylyzer validation report.

Finally, we verify if the migrated JP2 images are valid surrogates of the original TIF images by restoring the original TIF image from the converted JP2 image and comparing whether original and restored images are identical.

The sequential execution of this workflow is used as a reference point for measuring the parallelisation efficiency of the scalable version and it allows measuring how the processing times of the different components compare to each other.

The following diagram shows the average execution time of each component of the workflow in seconds and was created from a 1000 images sample of the Austrian National Library Tresor Music Collection:

 
!distribution_execution_times.PNG|border=1,width=473,height=264!

_Figure 2 (above): execution times of each of the workflows’ steps_

In the design phase this analysis is used to examine the average execution times for the individual tools. As a consequence of this experiment we might conclude, that over 4 seconds for the the FITS-based TIF image validation takes too much time and that this processing step needs to be improved, while the Jpylyzer validation is acceptable taking only slightly more than 1 second per image file in average.


h2. SCAPE Platform workflow - distributed processing

[Apache Pig|http://pig.apache.org/] was used to create a scalable version of this workflow. The different processing steps of the Taverna workflow for sequential processing are represented by Pig Latin statements.

The comments of each processing step In the script below indicate which is the corresponding processing component in the Taverna workflow.


{code}
/* file: tiff2jp2_migrate.pig */

/* Built from https://github.com/openplanets/tomar */
REGISTER /home/onbfue/ToMaR/target/tomar-1.5.2-SNAPSHOT.jar;

DEFINE ToMarService eu.scape_project.pt.udf.ControlLineUDF();
DEFINE XPathService eu.scape_project.pt.udf.XPathFunction();

SET job.name 'Tomar-Pig-Taverna-OpenJpeg';

/* make sure that one task per input file is created */
SET pig.noSplitCombination true;

SET mapred.task.timeout 420000

%DECLARE toolspecs_path '/hdfs/path/to/toolspecs';
%DECLARE xpath_exp1 '/fits/filestatus/valid';
%DECLARE xpath_exp2 '/fits/identification/identity/@mimetype';
%DECLARE xpath_exp3 '/jpylyzer/isValidJP2';

/* STEP 1: load image paths */
image_paths = LOAD '$image_paths' USING PigStorage() AS (image_path: chararray);

/* STEP 2: validation of tiff image files using fits  */
fits = FOREACH image_paths GENERATE image_path as image_path, ToMarService('$toolspecs_path', CONCAT(CONCAT('fits stdxml --input="hdfs://', image_path), '"')) as xml_text;

/* STEP 3: extract fits validity and mime-type using xpath */
fits_validation_list = FOREACH fits GENERATE image_path, XPathService('$xpath_exp1', xml_text) AS node_list1, XPathService('$xpath_exp2', xml_text) AS node_list2;
fits_validation = FOREACH fits_validation_list GENERATE image_path, FLATTEN(node_list1) as node1, FLATTEN(node_list2) as node2;
store fits into 'output/fits';
store fits_validation into 'output/fits_validation';

/* STEP 5: migration of tiff image files to jpeg2000 */
openjpeg = FOREACH fits_validation GENERATE image_path as image_path, ToMarService('$toolspecs_path',CONCAT( CONCAT(  CONCAT('openjpeg image-to-j2k --input="hdfs://', image_path), '" --output="'),  CONCAT(  CONCAT(   CONCAT('hdfs://', image_path), '.jp2'),'"'))) as ret_str;
STORE openjpeg INTO 'output/openjpeg';

/* STEP 6: validation of migrated jpeg2000 files using jpylyzer */
jpylyzer = FOREACH fits_validation GENERATE image_path as image_path, ToMarService('$toolspecs_path',CONCAT(CONCAT(CONCAT('jpylyzer validate --input="hdfs://', CONCAT(image_path,'.jp2')), '" --output="'),CONCAT(CONCAT( CONCAT('hdfs://', image_path), '.jp2.xml'),'"'))) as jpy_xml;
STORE jpylyzer INTO 'output/jpylyzer';

/* STEP 7: extract jpylyzer validity using xpath */
jpylyzer_validation_list = FOREACH jpylyzer GENERATE image_path, XPathService('$xpath_exp3', jpy_xml) AS jpy_node_list;
jpylyzer_validation = FOREACH jpylyzer_validation_list GENERATE image_path, FLATTEN(jpy_node_list) as node1;
store jpylyzer_validation into 'output/jpylyzer_validation';

/* STEP 8: migrate jpeg2000 image file back to tiff */
j2k_to_img = FOREACH fits_validation GENERATE image_path as image_path, ToMarService('$toolspecs_path',CONCAT( CONCAT(  CONCAT('openjpeg j2k-to-image --input="hdfs://', CONCAT(image_path,'.jp2')), '" --output="'),  CONCAT(  CONCAT(   CONCAT('hdfs://', image_path), '.jp2.tif'),'"'))) as j2k_to_img_ret_str;
STORE j2k_to_img INTO 'output/j2k_to_img';

/* STEP 9: compare orignal to restored image file */
imgcompare = FOREACH fits_validation GENERATE image_path as image_path, ToMarService('$toolspecs_path',CONCAT( CONCAT(CONCAT('imagemagick compare-pixelwise --inputfirst="hdfs://', image_path), CONCAT(CONCAT('" --inputsecond="hdfs://',CONCAT(image_path,'.jp2.tif')),'" --diffoutput="hdfs://')),CONCAT(image_path,'.cmp.txt"'))) as imgcompare_ret_str;
STORE imgcompare INTO 'output/imgcompare';
{code}


The following ToMaR tool specification files were used in this experiment:
* [^openjpeg.xml]
* [^imagemagick.xml] ([^compare.sh])
* [^fits.xml]
* [^jpylyzer.xml]

Note that these XML-based tool descriptions must be stored in the directory _/hdfs/path/to/toolspecs_ which is declared as the _toolspecs_path_ variable in the pig script above.


The script is then executed as follows:

{code}
pig -param image_paths=/hdfs/path/to/imagefiles/ tiff2jp2_migrate.pig
{code}
and produces the result files in the same directory where the input image files are located, for example, input image path /hdfs/path/to/imagefiles/imagefile.tif:
# /hdfs/path/to/imagefiles/imagefile.tif.jp2 (result of the conversion to JP2)
# /hdfs/path/to/imagefiles/imagefile.tif.jp2.tif (result of the re-conversion to TIF)
# /hdfs/path/to/imagefiles/imagefile.tif.txt (result of the pixel-wise comparison between original and re-converted TIF files)

h2. Evaluation summary

Files := Size of random sample

Total GB := Total size in Gigabytes

Secs := Processing time in seconds

Mins := Processing time in minutes

Hrs := Processing time in hours

Afg.p.f. := Average processing time per file in seconds

Obj/h := Number of objects processed per hour

GB/min := Throughput in Gigabytes per minute

GB/min := Throughput in Gigabytes per hour

Err := Number of processing errors


h3. Taverna Workflow - Sequential execution

| *Files* | *Total GB* | *Secs* | *Mins* | *Hrs* | *Avg.p.f.* | *Obj/h* | *GB/min* | *GB/h* | *Err* |
| 5 | 0,31 GB | 179 | 2,98 | 0,05 | 35,80 | 101 | 0,10 | 6,22 | 0 |
| 7 | 0,89 GB | 438 | 7,30 | 0,12 | 62,57 | 58 | 0,12 | 7,29 | 0 |
| 10 | 0,90 GB | 478 | 7,97 | 0,13 | 47,80 | 75 | 0,11 | 6,8 | 0 |
| 20 | 2,23 GB | 1150 | 19,17 | 0,32 | 57,50 | 63 | 0,12 | 6,98 | 0 |
| 30 | 2,99 GB | 1541 | 25,68 | 0,43 | 51,37 | 70 | 0,12 | 6,98 | 0 |
| 40 | 3,60 GB | 1900 | 31,67 | 0,53 | 47,50 | 76 | 0,11 | 6,81 | 0 |
| 50 | 3,46 GB | 2039 | 33,98 | 0,57 | 40,78 | 88 | 0,10 | 6,1 | 0 |
| 75 | 6,05 GB | 3425 | 57,08 | 0,95 | 45,67 | 79 | 0,11 | 6,36 | 0 |
| 100 | 8,30 GB | 4693 | 78,22 | 1,30 | 46,93 | 77 | 0,11 | 6,37 | 0 |
| 200 | 15,19 GB | 9246 | 154,10 | 2,57 | 46,23 | 78 | 0,10 | 5,91 | 0 |
| 300 | 19,07 GB | 11773 | 196,22 | 3,27 | 39,24 | 92 | 0,10 | 5,83 | 0 |
| 400 | 24,78 GB | 15644 | 260,73 | 4,35 | 39,11 | 92 | 0,10 | 5,70 | 0 |
| 500 | 34,55 GB | 21345 | 355,75 | 5,93 | 42,69 | 84 | 0,10 | 5,82 | 0 |
| 750 | 63,07 GB | 37397 | 623,28 | 10,39 | 49,86 | 72 | 0,10 | 6,07 | 0 |
| 1000 | 71,82 GB | 42376 | 706,27 | 11,77 | 42,38 | 85 | 0,10 | 6,10 | 0 |
| 2000 | 139,00 GB | 84938 | 1415,63 | 23,59 | 42,47 | 85 | 0,10 | 5,89 | 0 |
| 3000 | 211,85 GB | 128959 | 2149,32 | 35,82 | 42,99 | 84 | 0,10 | 5,91 | 0 |


h3. Pig Workflow - Distributed Execution

| *Files* | *Total GB* | *Secs* | *Mins* | *Hrs* | *Avg.p.f.* | *Obj/h* | *GB/min* | *GB/h* | *Err* |
| 5 | 0,31 GB | 96 | 1,60 | 0,03 | 19,20 | 188 | 0,19 | 11,60 | 0 |
| 7 | 0,89 GB | 101 | 1,68 | 0,03 | 14,43 | 250 | 0,53 | 31,64 | 0 |
| 10 | 0,90 GB | 103 | 1,72 | 0,03 | 10,30 | 350 | 0,53 | 31,56 | 0 |
| 20 | 2,23 GB | 114 | 1,90 | 0,03 | 5,70 | 632 | 1,17 | 70,45 | 0 |
| 30 | 2,99 GB | 138 | 2,30 | 0,04 | 4,60 | 783 | 1,30 | 77,99 | 0 |
| 40 | 3,60 GB | 161 | 2,68 | 0,04 | 4,03 | 894 | 1,34 | 80,41 | 0 |
| 50 | 3,46 GB | 183 | 3,05 | 0,05 | 3,66 | 984 | 1,13 | 68,01 | 0 |
| 75 | 6,05 GB | 272 | 4,53 | 0,08 | 3,63 | 993 | 1,34 | 80,11 | 0 |
| 100 | 8,30 GB | 373 | 6,22 | 0,10 | 3,73 | 965 | 1,34 | 80,15 | 0 |
| 200 | 15,19 GB | 669 | 11,15 | 0,19 | 3,35 | 1076 | 1,36 | 81,73 | 0 |
| 300 | 19,07 GB | 808 | 13,47 | 0,22 | 2,69 | 1337 | 1,42 | 84,95 | 0 |
| 400 | 24,78 GB | 1091 | 18,18 | 0,30 | 2,73 | 1320 | 1,36 | 81,77 | 0 |
| 500 | 34,55 GB | 1397 | 23,28 | 0,39 | 2,79 | 1288 | 1,48 | 89,03 | 0 |
| 750 | 63,07 GB | 2399 | 39,98 | 0,67 | 3,20 | 1125 | 1,58 | 94,64 | 0 |
| 1000 | 71,82 GB | 2746 | 45,77 | 0,76 | 2,75 | 1311 | 1,57 | 94,16 | 0 |
| 2000 | 139,00 GB | 5450 | 90,83 | 1,51 | 2,73 | 1321 | 1,53 | 91,82 | 0 |
| 3000 | 211,85 GB | 8328 | 138,80 | 2,31 | 2,78 | 1297 | 1,53 | 91,58 | 0 |
The following diagram shows the comparison of wall clock times in seconds (y-axis) of the Taverna workflow and the Pig workflow using an increasing number of files (x-axis).
!wallclocktime_concept_vs_scalable.PNG|border=1,width=648,height=425!

_Figure 3 (above): Wallclock times of concept workflow and scalable workflow_

However, the throughput we can reach using [this |SP:ONB Hadoop Platform]cluster and the chosen pig/hadoop job configuration is limited; as figure 4 shows, the throughput (measured in Gigabytes per hour -- GB/h) is rapidly growing when the number of files being processed is increased, and then stabilises at a value around slightly more than 90 Gigabytes per hour (GB/h) when processing more than 750 image files. !throughput_gb_per_h.png|border=1,width=654,height=363!
_Figure 4 (above): Throughput of the distributed execution measured in Gigabytes per hour (GB/h) against the number of files processed_