On Sun, 04 Sep 2016 15:41:58 -0400 Bill Cole wrote: > On 2 Sep 2016, at 17:38, RW wrote: > > [...] > > Running sa-compile gratuitously doesn't make much difference on most > > modern hardware, but it may do on something like a single-core > > Raspberry Pi. I remember on a single core Athlon, it was 30 minutes > > of 100% cpu usage. > > To do a sa-compile!?
That's what the clock said. > FWIW, it consistently takes less than a minute on the machine where I > do my rule QA and have the largest and most complex set of custom > rules in operation: a 2006 vintage 2GHz Core Duo (that's 32-bit) with > 2GB RAM and most of that otherwise in use. I'm not very surprised that a much more powerful cpu than the one I mentioned does it faster. I've no idea how fast it would be on something like a single-core Raspberry Pi, or other slow hardware. I'm just saying that, contrary to practically everything that's been written about this, you can skip running it after sa-update, if it's a problem. > Beyond it not really being very expensive to do, I'd think the main > advantage to always compiling the rules on larger systems would be in > loading the compiled shared object once for use by as many spamd > children or mimedefang slaves or whatever as you have running. That wrong in several ways, not least of which is that if you skip running sa-compile, the shared library is unaltered.
