I was microoptimising. I admit it. I was trying to get a little bit of code which read a part of a file going quickly, and I got slightly sidetracked. I noticed that PerlIO had layers, so I played around with the layers and benchmarked things. I noticed that :stdio was quite fast for my one test case, so I compiled a new Perl with ./Configure -des -Uuseperlio -Dprefix=/home/acme/perl-5.10.0-stdio/ to ignore PerlIO and use stdio instead, then benchmarked things (didn't really help). I eventually thought about using unbuffered IO, and noticed that it was much faster than all the others. Aha! I suddenly remember about buffered and unbuffered IO. Buffered IO using seek, read, and write is very handy when reading or writing the whole file. If you are actually doing random access in files then you should use sysseek, sysread and syswrite, doh!
