Richard, what we would like to have eventually is an option to select the
level of system quiescing. This will become very pertinent when we move on
to allowing selected memory objects to be snapshot - in both one or more
user spaces and system space. Whether or not the level of quiescing can be
automated according to context needs thinking about. From experience of
system dumping technologies on other operating systems, it does seem a
practical option to freeze by process in the case of user space. Given the
non-preemptable nature of the kernel it's not so easy to do this for kernel
space, but not impossible; it all depends on whether there is a user
context associated with a particular object and whether there is any
process level serialisation for such objects. A fully preemptable kernel
would make this much easier because serialisation would be much more
granular. Another option, which we used on other operating systems, is to
lock the kernel for a particular component, then yield before locking and
dumping the next component. Yes, you can get inconsistencies between the
data for each component, but that's not necessarily a bad thing.
Richard Moore - RAS Project Lead - Linux Technology Centre (ATS-PIC).
http://oss.software.ibm.com/developerworks/opensource/linux
Office: (+44) (0)1962-817072, Mobile: (+44) (0)7768-298183
IBM UK Ltd, MP135 Galileo Centre, Hursley Park, Winchester, SO21 2JN, UK
|