Right now in OpenSSI when we migrate a particular
application the entire allocated data with respect to
the application is taken to the node to which we are
migrating. That means if we have a 1GB of memory
allocated a lot of time will be spend in migration.
The above limitation/feature basically helps in
achieving the high availability. That means after
migration if the origin node goes down we still can
continue to run on the new node to which we migrated.
But i guess we should leave that to the user. How about
adding a /proc/partial_migration file that will help to
minimize the migration time in a OpenSSI cluster ? That
means pages are pulled from the origin node only when
we hit a page fault on the new node. This helps in
reducing the migration time a lot i guess. And in case
i move between nodes my net execution speed will be better.
Log in to post a comment.