>
Yes,
the same problem happens in the swap-over-NFS case: the nfs client
needs to allocate some memory (which is not disponible) , the the system
tries to
swap out pages via NFS, leading to an infinite recursion ...
I am not a kernel hacker, but I agree wich Luca thatthe RAID code
should include an option to allocate physical RAM (at startup ?) instead
of virtual mem,
so that it is able to do the necessary computations for the raid handling,
allowing
the swapping layer accessing to the MD device without conflicts.
one proposed swapping over RAID1:
I think it is much simpler to implement than over RAID5,
with RAID1 you can access the raw devices, like without raid,
with the difference that you must include code which writes to
n different disks the swap page, and in case of failure of a disk do the
similar
handling that the raid1 code does.
Anyone of the soft-RAID developers know if this issue is on their TODO list
?
best regards,
Benno.
>
> Benno,
> if i remember correctly the swap-on-raid problem was a memory starvation
> problem between the raid code needing memory for computations,
> so the system swaps out pages to disk, leading to the raid system
> needing memory, and so on. I don't believe that swapping to a file
> on an md device can do anithing to better this condition.
> I believe a solution might be reserving enough pages in kernel memory
> to handle raid, can this need be calculated????
>
> Regards,
> Luca
