> The answer is: you don't move the processes very often.
Right. But that is very different from not moving it. The use I see
for this, is to have a cluster in my basement that provides services,
including CPU. Now, in order to provide high availability, it is
necessary that processes can be migrated off a node being brought down
for maintenance. Similar for network connections, etc.
A node should also be able to maintain itself to some degree, perhaps
looking for indications of failure, and dissasociate itself from the
cluster, and shut down.
For single node setups, storing and restoring processes - e.g. making
them persist through power failure - could be important in many setups.
Most of this could be accomplished in user space, I'd think.
> The point is that if you move them at startup time, it's trivial to
> have the OS to do it and you can get any needed further load
> balancing out of small SMP boxes.
Similar, you can probably distribute incoming network connections to a
low-loaded machine.
> If I load balance statically at startup time and dynamically at
> runtime (within an SMP), what added performance will I get from moving
> already running processes? Answer: extremely little.
Well, if you load balance dynamically at startup time, and have long
running processes, you could be unfortunate with the distribution.
Eg. RC5 processes might benefit from migration.
But for performance, startup distribution is probably an almost-as-good
option, and, as you say, vastly simpler. It'd be nice if this were a
system issue, though, while there are parallelizing/distributed versions
of some software, they all require setting up, etc.
> If you are in the least work for the most bang mode, then you don't do
> process migration.
But this is Linux, we have an infinite number of monkeys :-)
> As I've repeatedly said, the dynamic changes are well handled by small SMPs.
Yes, well, okay. But I want to cluster a couple of old 486es...? Oh
well.
> As has been repeatedly proven, moving an already started process is a lose
> almost 100% of the time.
Right. Unless somebody just typed shutdown -hnow at the #-rompt.
~kzm
-- If I haven't seen further, it is by standing in the footprints of giants