It is recommended that you read through the High Performance Computing FAQ in addition to reading this FAQ.
Be aware that bowtie2 is preferred over bowtie, unless you have a reason to use the older bowtie. You can load either tool as a module. Use bowtie2 to map fastq sequence reads to an indexed reference genome or sequence. You can create your own reference index using bowtie2-build. If you need splice junction awareness in your mapping, use tophat instead of bowtie2. After mapping, you can use the cufflinks tools for analysis of differential expression or searching for novel genes/isoforms.
There are skeleton PBS submit files for running bowtie2/tophat/cufflinks in the HPC /genome folder. See /genome/bowtie_tophat.csh and /genome/tophat_HiMem.pbs. The latter uses a High Memory node and requests 80Gb of memory for tophat. There are a small number of HTC nodes with extra memory, so your job may have to wait until one of these "fat" nodes is available.