Joachim Schipper <[EMAIL PROTECTED]> writes: > Well, aside from the 'the load average is useless anyway' comments, a > load average of 1 is not necessarily unacceptable on a multi-processor > machine. (It means one CPU is constantly working.)
No, no, no and NO. Please. If you don't understand load averages, do not try to interpret them. It means that whenever the once a second event in the kernel happens, something was either in the run queue or the system made a guess that something might end up in the run queue within a second. That guess is not always correct. If you have something that wakes up once a second without doing anything, it will add to the load average. //art > > I understood that load averages were related to the number of processes > > in the run-queue (i.e. waiting to get some processor time) so average > > loads above 1 would show a processing bottleneck. > > > > Did I get this wrong? If so, what do the load averages show and what > > would be a high load for a dual CPU (4 core) system? > > That depends on a lot of things - most of which can not be read from the > load average - but as a data point, the system I'm typing this on is a > uniprocessor box. The CPU is 96% idle. The load average is 3 (backups > and checking out the OpenBSD tree from a local repository - i.e. the > bottleneck is the disk and the network, not the CPU). The system is > pretty responsive. > > However, if we are talking CPU-bound processes, anything less than 4 at > a time means that there is no (significant) contention for the bounding > resource, which roughly means they all run at full speed. This is not to > say that having some reserve capacity is not useful, or that you should > strive for 100% resource utilization, but running one CPU-bound process > is not a high load for this machine. (One disk-bound process, on the > other hand, may be.) > > Joachim > > -- > TFMotD: netstat (1) - show network status
