On Wed, October 8, 2008 17:27, Josh Luthman wrote:
Simply put the load is how much calculator power your box is doing. I have seen servers that have a value of several hundred load on a day to day basis.
Well, it depends on what sort of work the box is doing, and what your throughput expectations are, but any server at a load average of 200+ is going to have abysmal response times for online users. I've seen that on dedicated mail servers in a spam storm, but when our Oracle database servers go above about 30, we start getting calls from application users. The default alert levels in Hobbit work pretty well for us, letting us check whether anything's going awry or if it's a true spike in demand, before the phones start ringing. I do keep the dedicated Internet-facing sendmail servers at significantly higher thresholds, because mail doesn't necessarily have to move in near real time, and customers are more tolerant of slow delivery than "unable to deliver" warnings and bounces.