Andreas Metzler wrote:
> Hello,
> This is <http://bugs.debian.org/270735>. - I'll first quote the
> I am not sure what to make of this. Whether there is indeed a real
> problem and whether the ideas make sense at all.
>
> One of the obvious points is "reasonable value like 1.5 or 2". A
> machine with load 2 (e.g. building two software packages) still can
> receive and deliver mail without breaking into sweat, especially in
> the scenario the submitter was targeting, e-mail just being one of
> many services, i.e. rather light traffic.
>
If the machine is serving nothing but e-mail,
then there is no point in allowing the load to exceed about 2.
At that point, the processor and/or disk will be fully occupied,
and further increases in load will not result in any large
increase in throughput. For large values of load (e.g. 10),
you'll start to lose efficiency as the machine is doing
more context switches.
"Breaking into sweat" is a rather imprecise term.
I'm not 100% satisfied with my suggested solution,
but the problem of a hard limit is real. Hard limits
are only useful if you can be sure you're not DOSsing yourself.
A hard limit is therefore only useful if
there is some load value X such that
(a) your system never exceeds X for very long without considering
the load from e-mail, and
(b) X is small enough so that a mail blizzard won't slow
down everything else on your machine too severely.
(a) and (b) are going to depend strongly on how you use your system,
so you might not want to imagine that a typical system
does only e-mail, editing and occasional compilation.
I think it's a question of designing the software to be idiot-proof,
too. Hard limits are not idiot-proof; soft limits let the software
be misconfigured and still function. Since many of us (99.99% of
the computer users in the world) are not exim developers, it is
important to design software so that it can be mis-used.
