barton wrote:
Wouldn't it be reasonable and mostly trivial that
1) when a process is terminated, that any of it's pages that are on swap
are issued a CP
diagnose to discard the page? (requires a bit of Endicott work)
No. In case a process terminates, the memory is freed. For swap, this
does not imply any interaction to the underlying block device. Even if
the memory management would care, there currently is no call to do
that to a block device. A regular block driver (e.g. dasd) cannot do
any reasonable action in case such a call comes up.
Bottom line: changes to two different kernel subsystems, cross
platform code affected, only useful on s390. Useful but far from being
trivial.
2) when a page is swapped in, make that page anonymous again, and issue
that same
diagnose? This eliminates duplicate resources of expensive storage.
This would hurt performance a lot. The trick with sticky swap slots
is, that a page that is accessed read only in swap cache always stays
clean. Thus, it can be discarded again without the need to page out:
we do have an up-to-date swap slot out there. When swapping to regular
disks, and thus for 99% of Linux' users, this would be a significant
performance degradation because read accesses outnumber write accesses
in many pratical scenarios. Even with vdisk, when the guest is memory
constrained while CP is not, this could hurt performance overall.
cheers,
Carsten
----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
