Very interesting note for a new cluster checklist. Good to tune the file system down from 5%.
On a related note some operating systems ::cough:: FreeBSD will report negative disk space when you go over the quota. What does that mean? We run nagios with NRPE to run remote disk checks. We configure our alarms to warn at 5%. Image a process keeps using a disk. Alarm 5% , Alarm4%, Alarm 3%,Alarm 2%, Alarm 1%, Alarm cleared 103%. NRPE assumes disk % 0-100% comically if you drop below 100% the monitor thinks the disk is fine again. This is like a signed/unsigned bug. I mention this not to show off, As people deploy hadoop on other platforms issues like this might crop up. I am not saying it is an issue, but it could be.
