Easiest steps to recover unmapped reads from mapping?

I want to map raw Illumina reads to a rRNA database with millions of sequences. This has been recommended by a protocol online on Best Practices in de novo Assembly, to eliminate rRNA before assembly. These would be the unmapped reads in this step.

They recommend Bowtie2 which in Galaxy conveniently returns unmapped reads in a fasta file. However, I cannot complete this job because of memory restrictions – perhaps the database is too large. BWA is an option that works, and can be optimised for short illumina reads, however it does not return any fasta of unmapped reads.

Please, which would be the easiest solution? Either in making Bowtie2 work or getting unmapped reads in fasta from BWA? Any suggestions welcomed.

Thanks in advance!