Your subject line is a lie! According to the article you linked, Mr. Miller was a great guy.
But I see what you did there. You're just honoring roblimo by imitating him:
"Because Prodigy paid Miller based on user activity, he did his best to spark debate in the political chat room.
"I'd just say stuff to piss people off," Miller said"
Well it all depends on whether the Somali government will bail out the pirates. They'd better... I've got a CDS riding on the pirates not defaulting on this lady.
FTFY yvw
I'm guessing FreeBSD does this better than Linux.
So your httpds would probably be fine even on linux, but my server JVMs get pushed out and incur a big penalty to be paged back in.
I wouldn't advise people turn off swap either, just to be aware that the penalty is incurred, and that it can be avoided e.g. with a lower swappiness value and/or the mlock(2) call.
That is why many Unix admins recommend swap. It helps the system effectively manage its resources without incurring a penalty, so why wouldn't you?
Oh, but a big penalty is incurred. Always.
File buffers (for faster repeated reads) are a 100% total waste of RAM for a lot of disk reads that are never repeated, such as the nightly backup, slocate, makewhatis, and similar poltergoosts that come out between midnight and 4am localtime when nobody's looking.
But rather than completely turn off swap, I use vm.swappiness=20 on linux servers and workstations, much less swappy than the default of 60.
So to prevent the OS from swapping out my precious latency-sensitive BloatyApp (JVM) even if it's been idle for days, I recommend making a call to mlockall, through a JNI wrapper.
From mlock(2): "All mapped pages are guaranteed to be resident in RAM when the call returns successfully; the pages are guaranteed to stay in RAM until later unlocked."
My sister opened a computer store in Hawaii. She sells C shells down by the seashore.
