Statistics of the Common Crawl Corpus 2012
References
- Common Crawl Foundation
- The actual paper
- Raw index on AWS S3:
s3://aws-publicdatasets/common-crawl/index201
MapReduce code
The job run for creating the raw index is com.spiegler.fastindex.FastIndexer.java. It is a 'map only' job which outputs a single line entry for each website found in the CC corpus.
Each line contains the public suffix, domain, media type, charset, ARC file name and byte size of a specific website, all tab separated.
Build the job jar by running:
$ ant dist
to create dist/lib/Teneo-########.jar.
Run job on AWS
The job was run on 35 subsets of 25,000 ARC files of the 2012 corpus. Results were later merged into fewer files.
A job on a subset was invoked by
elastic-mapreduce --create --credentials credentials.json \
--jar s3://[bucket]/Teneo-########.jar \
--main-class com.spiegler.fastindex.FastIndexer \
--args "[AccessKey],[SecretKey],/home/hadoop/splits/split_1,s3://[bucket]/output/split_1" \
--instance-group master --instance-type m1.xlarge --instance-count 1 --bid-price [$$$] \
--instance-group core --instance-type m1.xlarge --instance-count 5 --bid-price [$$$] \
--bootstrap-action s3://elasticmapreduce/bootstrap-actions/install-ganglia \
--bootstrap-action s3://elasticmapreduce/bootstrap-actions/configurations/latest/memory-intensive \
--bootstrap-action s3://[bucket]/bootstrap_splits.sh \
--key-pair [YourKey] \
--log-uri s3n://[bucket] \
--enable-debugging
where the arguments for the job are the access key, secret key, a file containing the ARC file input list (bootstrapped onto instances) and an output S3 bucket.
The bootstrapping script bootstrap_splits.sh for copying split files onto the instances
#!/bin/bash
set -e
mkdir -p /home/hadoop/splits/
hadoop fs -copyToLocal s3://[bucket]/splits/* /home/hadoop/splits/
An example for a split, e.g. split_1
1346823845675/1346864466526_10.arc.gz
1346823845675/1346864469604_0.arc.gz
1346823845675/1346864469638_1.arc.gz
1346823845675/1346864471290_4.arc.gz
1346823845675/1346864477152_29.arc.gz
...
Hive code
For the actual aggregation Hive was used. Some examples are provided here.