Split a FASTQ (or pair) into 100K read splits using GNU split and pigz. Modified from an original script by @ekg.

fastq_splitter.sh

first_reads=$1
second_reads=$2

ddir=$(dirname $first_reads)
obase_first=$(basename $first_reads .fastq.gz)
obase_second=$(basename $second_reads .fastq.gz)

splitsz=4000000

if [ ! -z ${first_reads} ] && [ -e ${first_reads} ]
then
    time pigz -p4 -cd ${first_reads} | \
        split -d -a 6 -l ${splitsz} --filter='pigz -p4 > $FILE.gz' - ${ddir}/${obase_first}.fastq.part
else
    echo "ERROR: no file ${first_reads} found."
fi

if [ ! -z "${second_reads}" ] && [ -e ${second_reads} ]
then
    time pigz -p4 -cd ${second_reads} | \
    split -d -a 6 -l ${splitsz} --filter='pigz -p4 > $FILE.gz' - ${ddir}/${obase_second}.fastq.part
fi

If you want more parallelism (and you have the disk IO), you can combine this script with GNU parallel or LaunChair to parallelize by fastq file as well.

GNU parallel on all fastqs in a directory, using four jobs (e.g. for a 16 core system):

ls | grep ".fastq.gz$" | parallel -j 4 './fastq_splitter.sh {}'

LaunChair example on a 16-core system:

git clone --recursive https://github.com/edawson/LaunChair

for i in `ls | grep -v "ERR894723" | grep "fastq.gz"`;
do 
   echo $i; echo "./fq_splitter.sh $i" >> jobfile.txt
done
python LaunChair/launcher.py -i jobfile.txt -c 4 -n 16