|
> More on the LVS scheduling Honours Project / Thesis
>
>
> Oh I forgot to mention, I'll also be doing simulations and testing and
> comparisons between the existing load balancing algorithms and also
> between them and the new code that I (hopefully) come up with.
(This may already have been discussed but......)
An idea:
Create hooks, probably writeable files in /proc, to allow a user space
tool to inform the LVS kernel code of the "desirability" of each server.
You could then write a single kernel level scheduler for LVS (called
"lvs_user_schedule") that uses this info in it's scheduling decisions.
This would let you push the more complicated portions of your fuzzy logic
scheduler into user space (and possibly help in maintaining your own
sanity :).
|
